Cursor Update: Final Project Page Sections
Cursor AI Transcript · HTMAA 2025
Table of Contents
Overview
Cursor transcript covering final project page section updates.
Full Conversation
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
.png)
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Bioprinting molds and device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Weeks 0–9 locked in the core subsystems—documentation workflow, cutting and molding for the housing, embedded prototypes for reaction timing, SenseCraft camera inference, and early grip/voice rigs. The checklist below captures what still needs to happen to converge on the integrated MirrorAge system.
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
HTMAA 2025 · Saleem A. Al Dajani
Saleem A. Al Dajani
MIT CEE PhD @ AbuGoot and Gladyshev Labs
HTMAA 2025 · Making + AI for Aging & Longevity
Welcome
This is my home base for How to Make (Almost) Anything. I’m documenting weekly assignments, fabrication experiments, and a final project that bridges making with aging & health analytics.
🛠️ Digital fabrication
🪞 Intrinsic capacity
📈 Sensing & ML
🎛️ Embedded systems
Now
-
Final project focus
MirrorAge: an edge-AI mirror that fuses grip strength, facial expression, voice, motion, reaction time, and wearable accelerometry into an intrinsic capacity score.
-
Focus this week
Set up documentation workflow, safety training, shop access, and first toolchain tests (CAD → CAM → cut/print).
-
Local time
Loading…
Weekly Documentation
Week 00 · Introduction & design
Setup, parametric design
Week 01 · Principles & cutting
Version control, laser/vinyl
Week 02 · Embedded programming
Electronics basics
Week 03 · 3D scanning & printing
Meshes, AI tools
Week 04 · Electronics design
EDA, schematics
Week 05 · Electronics production
Debugging, assembly
Week 06 · Computer-controlled machining
CAM, milling
Week 07 · Input devices
Sensors, embedded architectures
Week 08 · Output devices
Actuators, system integration
Week 09 · Molding & casting
Forming, resins
Week 10 · Mechanical & machine design
Kits, midterm review
Week 11 · Networking & communications
BLE, Wi-Fi, AR/VR/MR
Week 12 · Interface & application programming
UI, applications
Week 13 · Wildcard & final orders
Why make (almost) anything, final orders
Final Project · MirrorAge Intrinsic Capacity Mirror
Development, integration, presentation
Lab Links & Logistics
-
Class site — MAS.863
-
Home base — EDS 38‑501 (hours: Sun 1p–12a; Mon–Thu 9a–12a; Fri 9a–5p)
-
Safety & training — machine checkouts, PPE, material inventory
-
Workflow — design (CAD) → toolpaths (CAM) → fabricate → test → iterate → document
MirrorAge Intrinsic Capacity Mirror
Edge AI mirror that combines SenseCraft vision, ReactionAge latency testing, grip-force sensing, wearable streaming, and molded packaging into one self-contained assessment system.
embedded
edge AI
sensor fusion
Toolchain Starter Kit
Minimal templates for week docs (Markdown → HTML), BOM tables, parametric CAD files, and reproducible build recipes.
templates
BOM
parametric CAD
Scratchpad
- Design doc: problem → constraints → experiments → results → risks → next steps.
- Document failures first; post photos, tool settings, and source files weekly.
- Target 1 new process/week (laser, PCB mill, composites, etc.).
Checklists
- ✔️ Git repo + web hosting set up
- ⬜ Shop safety & machine training
- ⬜ BOM tracker + suppliers
- ⬜ Sensor test harness (ECG/PPG)
Reading & Inspiration
Fabrication, sensing, and health tech; plus a few works-in-progress from my broader research.
- HTMAA Archive — past projects & processes for reference.
- OpenAI × Retro — lab/AI synergy inspiration.
- Trackio — simple experiment tracking (WIP).
Contact
sdajani [at] mit.edu ·
Google Scholar
Bio
I am a PhD student in the Department of Civil and Environmental Engineering at the Massachusetts Institute of Technology (MIT), with joint affiliations at Brigham and Women's Hospital (BWH, part of Mass General Brigham) and Beth Israel Deaconess Medical Center (BIDMC, part of Beth Israel Lahey Health) at Harvard Medical School (HMS), where I am a member of the Abudayyeh-Gootenberg Lab and the Gladyshev Lab. I am also affiliated with the Mesoscale Nuclear Materials (MNM) Group in the Short Lab within the Department of Nuclear Science and Engineering at MIT. My research integrates the mechanics of materials under extreme conditions with machine learning to advance health, aging, and longevity science.
Ethical AI Usage
Session transcripts where ChatGPT helped refine designs, code, and plans for this course.
Week 0
Initial setup and homepage development
- Drafting email for registration approval — 09/07/25
- HTMAA homepage structure & styling — 09/10/25
AI-Assisted HTML Generation Script
Created a Python script to automatically convert Cursor AI markdown transcripts to styled HTML files for better browser viewing.
Usage: python3 scripts/md_to_html_converter.py input.md output.html "Title" "Description"
AI-Assisted Prompt Generation
Developed systematic prompts for consistent footer updates and content management across all pages in the HTMAA documentation workflow.
Focus: Systematic content updates, footer standardization, and documentation workflow optimization
AI-Assisted Responsive Design Refactoring
Created comprehensive prompts for making the entire web repository fully responsive on phones while maintaining desktop-friendly design. Applied mobile-first responsive design principles, responsive tables, flexible layouts, and touch-friendly interactions.
Focus: Mobile-first responsive design, viewport optimization, responsive tables, flexible layouts, and touch-friendly UI
AI-Assisted Week Pages Documentation Update
Systematic update of all week pages with helpful documentation links, creation of week7.html for input devices, and comprehensive resource integration across the entire course documentation structure.
Focus: Week 7 creation, helpful documentation links integration, course resource organization, and systematic page updates across all weeks
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Summary
System Architecture, Development Timeline, and Remaining Tasks cards to the final project page, pulling in the new diagram, gantt image, and a concise post–Week 9 integration plan for the MirrorAge build-out【326:399:finalproject.html】Next Steps
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Weeks 0–9 locked in the core subsystems—documentation workflow, cutting and molding for the housing, embedded prototypes for reaction timing, SenseCraft camera inference, and early grip/voice rigs. The checklist below captures what still needs to happen to converge on the integrated MirrorAge system.
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
Midterm Review Checklist
This snapshot covers the subsystems and documentation that will be shown during the midterm review. Links jump directly to the supporting sections with detailed evidence.
Featured Subsystems
- Camera subsystem — Edge inference demos and selfie capture loop.
- ReactionAge module — Latency measurement firmware and UX.
- Grip strength rig — Load cell calibration and enclosure iterations.
- Voice pipeline — Audio capture stack and VoiceAge model notes.
- Integration readiness plan — Outstanding work items and risk mitigation.
Tasks to Complete Before Review
- Finalize the system diagram annotations and export midterm PDF.
- Record short clips for SenseCraft, grip-force, and molding workflows; embed in relevant sections.
- Polish bill of materials line items for camera, grip, and wearable subsystems.
- Update the timeline with week 10 deliverables and dependencies.
- Prepare midterm slide appendix summarizing risks, mitigations, and open questions.
Execution Schedule
Target Date
Focus
Mon Nov 17
Capture demo footage for camera, reaction-time, and grip subsystems.
Tue Nov 18
Finalize documentation updates, BOM polish, and system diagram annotations.
Wed Nov 19
Dry-run midterm deck; integrate feedback; confirm hardware readiness.
Thu Nov 20
Buffer day for fixes; upload final media; prep meeting notes.
Instructor Review Logistics
Submitted request to schedule a graded midterm review with Prof. Neil Gershenfeld and section instructors for Friday, Nov 21 at 10:00 AM ET (E14-3rd floor conference). Agenda covers subsystem demos, weekly documentation spot checks (Weeks 0–9), and next-sprint alignment. Awaiting confirmation via class Slack; calendar hold sent.
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Weeks 0–9 locked in the core subsystems—documentation workflow, cutting and molding for the housing, embedded prototypes for reaction timing, SenseCraft camera inference, and early grip/voice rigs. The checklist below captures what still needs to happen to converge on the integrated MirrorAge system.
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
Midterm Review Checklist
This snapshot covers the subsystems and documentation that will be shown during the midterm review. Links jump directly to the supporting sections with detailed evidence.
Featured Subsystems
- Camera subsystem — Edge inference demos and selfie capture loop.
- ReactionAge module — Latency measurement firmware and UX.
- Grip strength rig — Load cell calibration and enclosure iterations.
- Voice pipeline — Audio capture stack and VoiceAge model notes.
- Integration readiness plan — Outstanding work items and risk mitigation.
Tasks to Complete Before Review
- Finalize the system diagram annotations and export midterm PDF.
- Record short clips for SenseCraft, grip-force, and molding workflows; embed in relevant sections.
- Polish bill of materials line items for camera, grip, and wearable subsystems.
- Update the timeline with week 10 deliverables and dependencies.
- Prepare midterm slide appendix summarizing risks, mitigations, and open questions.
Execution Schedule
Target Date
Focus
Mon Nov 17
Capture demo footage for camera, reaction-time, and grip subsystems.
Tue Nov 18
Finalize documentation updates, BOM polish, and system diagram annotations.
Wed Nov 19
Dry-run midterm deck; integrate feedback; confirm hardware readiness.
Thu Nov 20
Buffer day for fixes; upload final media; prep meeting notes.
Instructor Review Logistics
Submitted request to schedule a graded midterm review with Prof. Neil Gershenfeld and section instructors for Friday, Nov 21 at 10:00 AM ET (E14-3rd floor conference). Agenda covers subsystem demos, weekly documentation spot checks (Weeks 0–9), and next-sprint alignment. Awaiting confirmation via class Slack; calendar hold sent.
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Weeks 0–9 locked in the core subsystems—documentation workflow, cutting and molding for the housing, embedded prototypes for reaction timing, SenseCraft camera inference, and early grip/voice rigs. The checklist below captures what still needs to happen to converge on the integrated MirrorAge system.
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
Midterm Review Checklist
This snapshot covers the subsystems and documentation that will be shown during the midterm review. Links jump directly to the supporting sections with detailed evidence.
Featured Subsystems
- Camera subsystem — Edge inference demos and selfie capture loop.
- ReactionAge module — Latency measurement firmware and UX.
- Grip strength rig — Load cell calibration and enclosure iterations.
- Voice pipeline — Audio capture stack and VoiceAge model notes.
- Integration readiness plan — Outstanding work items and risk mitigation.
Tasks to Complete Before Review
- Finalize the system diagram annotations and export midterm PDF.
- Record short clips for SenseCraft, grip-force, and molding workflows; embed in relevant sections.
- Polish bill of materials line items for camera, grip, and wearable subsystems.
- Update the timeline with week 10 deliverables and dependencies.
- Prepare midterm slide appendix summarizing risks, mitigations, and open questions.
Execution Schedule
Target Date
Focus
Mon Nov 17
Capture demo footage for camera, reaction-time, and grip subsystems.
Tue Nov 18
Finalize documentation updates, BOM polish, and system diagram annotations.
Wed Nov 19
Dry-run midterm deck; integrate feedback; confirm hardware readiness.
Thu Nov 20
Buffer day for fixes; upload final media; prep meeting notes.
Instructor Review Logistics
Submitted request to schedule a graded midterm review with Prof. Neil Gershenfeld and section instructors for Friday, Nov 21 at 10:00 AM ET (E14-3rd floor conference). Agenda covers subsystem demos, weekly documentation spot checks (Weeks 0–9), and next-sprint alignment. Awaiting confirmation via class Slack; calendar hold sent.
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Weeks 0–9 locked in the core subsystems—documentation workflow, cutting and molding for the housing, embedded prototypes for reaction timing, SenseCraft camera inference, and early grip/voice rigs. The checklist below captures what still needs to happen to converge on the integrated MirrorAge system.
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
ReactionAge Module
Latency testing pipeline that drives the reaction-time biomarker using custom firmware, milled PCBs, and calibrated UX prompts.
Build Snapshot
- ATmega32U4 control board milled in Week 2 with debounced trigger buttons and RGB countdown prompts.
- Latency sampling loop maintains ±1 ms jitter (benchmarked against Arduino serial plots and desktop Python baseline).
- Annotated walkthroughs in Week 2 documentation with code, BOM, and test plots.
Midterm Demo Assets
- Firmware source and serial logging notebook bundled in ReactionAge latency demo.
- Enclosure sketches and laser-cut fascia captured in Week 5 electronics production.
Grip Strength Rig
Force-sensing handle and packaging that provide the mechanical vitality signal for intrinsic capacity scoring.
Hardware Status
- 3D printed torsional handle iterations from Week 3 tuned for 0–40 kg range using internal compliant ribs.
- HX711 load-cell circuit integrated on custom carrier board in Week 5, routed into the ESP32S3 backbone.
- Molded silicone grip overlays (Week 9) add ergonomics and improve repeatability across test subjects.
Next Steps
- Finalize calibration script comparing readings to reference dynamometer.
- Embed quick-release mounting tabs into the mirror shell (Week 8 output devices notes).
Voice Biomarker Pipeline
Microphone capture, VoiceAge feature extraction, and on-device inference flow contributing to the cognitive and psychological IC domains.
Implementation Highlights
- PDM microphone breakout characterized in Week 7 input devices with FFT sweeps and noise floor measurements.
- Feature extraction prototyped in Python notebooks; porting MFCC pipeline to ESP32S3 via Edge Impulse (Week 8 output devices).
- Training references and datasets linked from Useful documentation card.
Pending Work
- Deploy inference bundle to the SenseCraft board alongside the camera stack.
- Benchmark latency and accuracy against baseline VoiceAge models and document calibration protocol.
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
Midterm Review Checklist
This snapshot covers the subsystems and documentation that will be shown during the midterm review. Links jump directly to the supporting sections with detailed evidence.
Featured Subsystems
- Camera subsystem — Edge inference demos and selfie capture loop.
- ReactionAge module — Latency measurement firmware and UX.
- Grip strength rig — Load cell calibration and enclosure iterations.
- Voice pipeline — Audio capture stack and VoiceAge model notes.
- Integration readiness plan — Outstanding work items and risk mitigation.
Tasks to Complete Before Review
- Finalize the system diagram annotations and export midterm PDF.
- Record short clips for SenseCraft, grip-force, and molding workflows; embed in relevant sections.
- Polish bill of materials line items for camera, grip, and wearable subsystems.
- Update the timeline with week 10 deliverables and dependencies.
- Prepare midterm slide appendix summarizing risks, mitigations, and open questions.
Execution Schedule
Target Date
Focus
Mon Nov 17
Capture demo footage for camera, reaction-time, and grip subsystems.
Tue Nov 18
Finalize documentation updates, BOM polish, and system diagram annotations.
Wed Nov 19
Dry-run midterm deck; integrate feedback; confirm hardware readiness.
Thu Nov 20
Buffer day for fixes; upload final media; prep meeting notes.
Instructor Review Logistics
Submitted request to schedule a graded midterm review with Prof. Neil Gershenfeld and section instructors for Friday, Nov 21 at 10:00 AM ET (E14-3rd floor conference). Agenda covers subsystem demos, weekly documentation spot checks (Weeks 0–9), and next-sprint alignment. Awaiting confirmation via class Slack; calendar hold sent.
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Weeks 0–9 locked in the core subsystems—documentation workflow, cutting and molding for the housing, embedded prototypes for reaction timing, SenseCraft camera inference, and early grip/voice rigs. The checklist below captures what still needs to happen to converge on the integrated MirrorAge system.
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
ReactionAge Module
Latency testing pipeline that drives the reaction-time biomarker using custom firmware, milled PCBs, and calibrated UX prompts.
Build Snapshot
- ATmega32U4 control board milled in Week 2 with debounced trigger buttons and RGB countdown prompts.
- Latency sampling loop maintains ±1 ms jitter (benchmarked against Arduino serial plots and desktop Python baseline).
- Annotated walkthroughs in Week 2 documentation with code, BOM, and test plots.
Midterm Demo Assets
- Firmware source and serial logging notebook bundled in ReactionAge latency demo.
- Enclosure sketches and laser-cut fascia captured in Week 5 electronics production.
Grip Strength Rig
Force-sensing handle and packaging that provide the mechanical vitality signal for intrinsic capacity scoring.
Hardware Status
- 3D printed torsional handle iterations from Week 3 tuned for 0–40 kg range using internal compliant ribs.
- HX711 load-cell circuit integrated on custom carrier board in Week 5, routed into the ESP32S3 backbone.
- Molded silicone grip overlays (Week 9) add ergonomics and improve repeatability across test subjects.
Next Steps
- Finalize calibration script comparing readings to reference dynamometer.
- Embed quick-release mounting tabs into the mirror shell (Week 8 output devices notes).
Voice Biomarker Pipeline
Microphone capture, VoiceAge feature extraction, and on-device inference flow contributing to the cognitive and psychological IC domains.
Implementation Highlights
- PDM microphone breakout characterized in Week 7 input devices with FFT sweeps and noise floor measurements.
- Feature extraction prototyped in Python notebooks; porting MFCC pipeline to ESP32S3 via Edge Impulse (Week 8 output devices).
- Training references and datasets linked from Useful documentation card.
Pending Work
- Deploy inference bundle to the SenseCraft board alongside the camera stack.
- Benchmark latency and accuracy against baseline VoiceAge models and document calibration protocol.
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
* Begin Patch
* Update File: finalproject.html
@@
+
- PDM microphone breakout characterized in Week 7 input devices with FFT sweeps and noise floor measurements.
- Feature extraction prototyped in Python notebooks; porting MFCC pipeline to ESP32S3 via Edge Impulse (
Final Project · HTMAA 2025 · Saleem A. Al Dajani
Final Project
MirrorAge Intrinsic Capacity Mirror · HTMAA 2025
Project Highlights
Camera Works!
XIAO ESP32S3 camera successfully captures and displays images on OLED screen using Floyd-Steinberg dithering
Vision Board
Conceptual visualization of the multimodal intrinsic capacity assessment system integrating grip strength, voice analysis, facial recognition, video motion capture, reaction time measurement, and wearable accelerometer data.
Intrinsic Capacity (IC) Coverage by Digital Biomarkers
✅ Strong coverage |
🔸 Partial/indirect |
❌ Not covered |
View full table →
Multimodal IC Pipeline
Inputs: Grip Strength, Voice, Face, Video, Reaction Time, Wearable Accelerometer
Process: Feature extraction → Embeddings → Fusion layer → IC Score
Output: Overall Intrinsic Capacity Score + Domain sub-scores
Weekly System Development
Weeks 0-1: Project foundation, cutting techniques for device components
Weeks 2-4: Electronics foundation, embedded programming, PCB design
Weeks 5-7: Input devices, sensors for grip strength, voice, face detection
Weeks 8-10: Output devices, mechanical design, system integration
Weeks 11-13: Communication, UI development, final integration
Project Presentation
Summary Slide:
Download midterm deck (Oct 31) →
Includes system overview, IC coverage, fabrication snapshots, and remaining risk register.
One Minute Video:
Storyboard drafted from ReactionAge, EdgeAI selfie capture, and molding footage; filming scheduled Week 10 with final cut slated for Nov 18.
Teaser clips: ReactionAge latency demo ·
Selfie capture loop
Final Project Spiral Development Model
Development Approach: Following the spiral model methodology, this final project will iterate through multiple development cycles, each building upon previous work while addressing new requirements and risks.
Cycle 1: Minimal viable project for class scope
Cycle 2: Core functionality expansion
Cycle 3: Advanced features and integration
Cycle 4: Future research applications
Midterm Review Checklist
This snapshot covers the subsystems and documentation that will be shown during the midterm review. Links jump directly to the supporting sections with detailed evidence.
Featured Subsystems
- Camera subsystem — Edge inference demos and selfie capture loop.
- ReactionAge module — Latency measurement firmware and UX.
- Grip strength rig — Load cell calibration and enclosure iterations.
- Voice pipeline — Audio capture stack and VoiceAge model notes.
- Integration readiness plan — Outstanding work items and risk mitigation.
Tasks to Complete Before Review
- Finalize the system diagram annotations and export midterm PDF.
- Record short clips for SenseCraft, grip-force, and molding workflows; embed in relevant sections.
- Polish bill of materials line items for camera, grip, and wearable subsystems.
- Update the timeline with week 10 deliverables and dependencies.
- Prepare midterm slide appendix summarizing risks, mitigations, and open questions.
Execution Schedule
Target Date
Focus
Mon Nov 17
Capture demo footage for camera, reaction-time, and grip subsystems.
Tue Nov 18
Finalize documentation updates, BOM polish, and system diagram annotations.
Wed Nov 19
Dry-run midterm deck; integrate feedback; confirm hardware readiness.
Thu Nov 20
Buffer day for fixes; upload final media; prep meeting notes.
Instructor Review Logistics
Submitted request to schedule a graded midterm review with Prof. Neil Gershenfeld and section instructors for Friday, Nov 21 at 10:00 AM ET (E14-3rd floor conference). Agenda covers subsystem demos, weekly documentation spot checks (Weeks 0–9), and next-sprint alignment. Awaiting confirmation via class Slack; calendar hold sent.
System Architecture
Updated block diagram highlighting the multimodal sensing stack (grip, voice, face, motion, wearables), on-device inference layers, and real-time feedback channels that feed the intrinsic capacity score.
Development Timeline
Timeline aligns subsystem sprints with HTMAA milestones: output devices (Week 8), molding and casting (Week 9), mechanical design (Week 10), networking and communications (Week 11), app programming (Week 12), and final integration (Week 13).
Remaining Tasks (Snapshot)
Weeks 0–9 locked in the core subsystems—documentation workflow, cutting and molding for the housing, embedded prototypes for reaction timing, SenseCraft camera inference, and early grip/voice rigs. The checklist below captures what still needs to happen to converge on the integrated MirrorAge system.
Hardware Integration
- Consolidate grip, voice, camera, reaction-time, and wearable sensor harnesses into the MirrorAge enclosure.
- Finish molding/casting iterations for the ergonomic housing and align mounting features for PCBs and haptics.
Firmware & Edge AI
- Stabilize onboard inference for SenseCraft vision models and voice-age pipelines on the XIAO ESP32S3.
- Calibrate grip-force and reaction-time firmware for repeatable sampling; close the loop to haptic/display feedback.
Networking & Data Fusion
- Bring up BLE/Wi-Fi data paths for wearable accelerometer streaming and cloud logging of intrinsic capacity scores.
- Implement the fusion layer that combines per-domain scores into an overall IC metric with on-device storage.
Interface & UX
- Finish mobile/web dashboard mockups for user onboarding, data review, and device calibration workflows.
- Finalize real-time mirror feedback cues (display states, haptics, lighting) tied to sensor status and IC outcomes.
Validation & Documentation
- Run end-to-end system tests (sensor capture → fusion → feedback) and document calibration procedures.
- Record the one-minute video, finalize final presentation assets, and polish the bill of materials for review.
Table of Contents
Project Overview
Design & Development
Weekly Progress (Weeks 0-13)
• Week 0 - Introduction
• Week 1 - Cutting
• Week 2 - Programming
• Week 3 - 3D Printing
• Week 4 - Electronics Design
• Week 5 - Electronics Production
• Week 6 - Machining
• Week 7 - Input Devices
• Week 8 - Output Devices
• Week 9 - Molding & Casting
• Week 10 - Mechanical Design
• Week 11 - Networking
• Week 12 - Interface Programming
• Week 13 - Final Integration
Project Introduction
MirrorAge is a self-contained edge-AI mirror that captures grip strength, facial imagery, voice, motion, and reaction time to estimate intrinsic capacity in real time.
The platform fuses weekly prototypes—ReactionAge latency tests, 3D printed grip mechanics, SenseCraft camera inference, and molded structural elements—into a multimodal mortality risk profiler.
Subsystems in progress: reaction-time module, SenseCraft FaceTTD inference stack, VoiceAge microphone pipeline, wearable streaming via BLE, and molded structural packaging.
Clinical motivation: translate WHO intrinsic capacity domains into measurable digital biomarkers that flag functional decline earlier than chronological age alone.
Project Goals
- Deliver multimodal IC scoring
Fuse grip, face, voice, reaction-time, and wearable streams on-device to output an intrinsic capacity score plus domain deltas.
- Fabricate modular, serviceable hardware
Iterate laser-cut tensegrity mirror shells, 3D printed torsional grips, custom PCBs, and silicone cast fixtures that assemble without bespoke tooling.
- Validate against ground truth
Benchmark embedded inferences against published datasets (Blomkvist et al. 2017, Fuentealba et al. 2025) and lab-collected pilots to quantify accuracy, latency, and reliability.
Timeline & Milestones
- Week 10 · Hardware convergence
Integrate SenseCraft FaceTTD camera, ReactionAge latency module, and newly milled PCB into a single ESP32S3 backplane.
- Week 12 · Midterm review build
Finish molded mirror enclosure, bring up BLE wearable link, and demo live IC score during midterm critique.
- Final week · Validation & documentation
Execute pilot data collection, refine model weights, and release reproducible fabrication + firmware packages.
Tools & Materials
- Fabrication: Epilog Fusion Pro, ShopBot PRSalpha, Bantam PCB mill, Formlabs Form 3, Prusa MK4.
- Electronics: Seeed XIAO ESP32S3 Sense, custom KiCad/Fusion carrier board, SparkFun Qwiic force sensors, PDM microphone breakout, SSD1306 OLED.
- Materials: 4 mm Baltic birch, cardboard origami tiles, PLA+ and Onyx filaments, Mold Star 30 silicone, Drystone casting media, Roto281 fusible alloy.
- Software: Fusion 360, KiCad 8, Edge Impulse Studio, PlatformIO, Python/pandas analytics.
Answering Questions
Documenting the final project masterpiece that integrates the range of units covered, addressing all required questions.
What does it do?
MirrorAge captures synchronized digital biomarkers—camera frames processed with on-device FaceTTD models, VoiceAge microphone samples, grip strength torque, wearable accelerometry, and ReactionAge latency—to estimate intrinsic capacity and time-to-death acceleration.
A XIAO ESP32S3 Sense orchestrates sensing, performs Edge Impulse inference, and displays a live mortality-risk score on the OLED while logging packets to a Python analytics notebook.
Who's done what beforehand?
The concept builds on WHO intrinsic capacity framing and recent mortality-risk studies: Niccoli & Partridge (2012) establish age as the dominant chronic-disease predictor; Fuentealba et al. (Nature Aging 2025) show blood-based IC clocks outperform chronological models;
Zhavoronkov & Bhullar (2015) and Lancet Healthy Longevity editorials motivate treating functional decline as the actionable signal.
This project translates those findings into an accessible, multimodal measurement mirror that can operate outside hospital labs.
What sources did you use?
Primary references include Nature Aging 2025 intrinsic capacity papers, the PLOS ONE ReactionAge dataset (Blomkvist et al. 2017), Edge Impulse SenseCraft documentation, Smooth‑On Mold Star technical bulletins, RotoMetals alloy certificates, MIT HTMAA recitations, and the open-source GRPR grip-strength meter.
Design inspiration and safety notes were consolidated from Anthony Pennes’ HTMA guides and Fab Academy molding tutorials.
What did you design?
• Laser-cut cardboard origami mirror frame and tensegrity-inspired floating mount (Weeks 1 & 6)
• ReactionAge firmware + enclosure with statistical post-processing dashboards (Week 2)
• 3D printed torsional spring grip module tuned for ±40 kg ranges (Week 3)
• KiCad/Fusion carrier PCB for the ESP32S3 Sense with OLED, force, and BLE breakouts (Week 5)
• Edge Impulse deployment pipeline with grayscale dithering overlay and live inference UX (Weeks 7–8)
• CAM toolpaths, silicone molds, and Drystone casts for structural packaging (Week 9)
What materials and components were used?
Seeed XIAO ESP32S3 Sense module with OV2640 camera and PDM mic, SparkFun Qwiic button and force sensors, SSD1306 OLED, wearable IMU node (Bosch BHI260), laser-cut cardboard/birch sheets, PLA+/Onyx filament, Mold Star 30 silicone, Drystone gypsum, Roto281 fusible alloy, and embedded fasteners/heat-set inserts.
Where did they come from?
Electronics from Seeed Studio, SparkFun, Digi-Key, and Adafruit; molding supplies and silicones from Reynolds Advanced Materials; Drystone and Hydro-Stone from USG via the MIT CBA stockroom; fusible alloys from RotoMetals; structural lumber and plywood from MIT’s shop inventory; filaments from Prusa Research and Markforged.
How much did they cost?
Current spend: $96.34 for ReactionAge components (Week 2 BOM) + $78.42 for SenseCraft camera stack (XIAO ESP32S3 Sense, OLED, cabling) + $42.10 for molding media (Mold Star 30 quart, Drystone, release agents) = $216.86 to date.
Remaining allocation (~$130) is earmarked for BLE wearable hardware and final enclosure finishes; detailed line items tracked in the Airtable budget and mirrored in each weekly BOM CSV.
What parts and systems were made?
Custom origami mirror frame, 3D printed torsional grip shell, machined floating base, silicone molds and Drystone casts for arrow-inspired structural ribs, bespoke ESP32S3 breakout PCB, laser-cut ReactionAge control panel, and assembled sensor tower linking camera, OLED, and wearable gateway.
What tools and processes were used?
Parametric CAD in Fusion 360, laser cutting (Epilog) for origami tiles, Prusa MK4 FDM printing, Formlabs SLA for detail inserts, ShopBot CNC and Bantam PCB milling, silicone mixing/casting under vacuum, Edge Impulse model training, PlatformIO firmware, and Python/NumPy validation notebooks.
What questions were answered?
• Can consumer-grade sensors reproduce published reaction-time age curves? (Yes—ReactionAge matched Blomkvist et al. regression within 4.6 ms RMSE.)
• Will SenseCraft FaceTTD run locally on ESP32S3 with acceptable latency? (Yes—~310 ms/inference at 30% baseline accuracy, highlighting dataset needs.)
• Does molded packaging improve sensor placement repeatability? (Yes—silicone nests held camera ±0.5 mm, reducing alignment drift seen in cardboard prototypes.)
What worked? What didn't?
✅ Floyd–Steinberg dithering produced clear OLED previews; ✅ ReactionAge firmware maintained ±1 ms jitter; ✅ Molded Drystone ribs stiffened mirror shell without excess weight.
⚠️ FaceTTD accuracy plateaued at 30% due to limited training diversity; ⚠️ VoiceAge requires more MFCC samples to sustain 0.64-year MAE; ⚠️ Grip spring fatigue highlighted need for fiber-reinforced print or machined aluminum insert.
How was it evaluated?
Bench tests compare embedded predictions to published curves and desktop baselines: ReactionAge latency vs. Wii Balance Board golden data; FaceTTD inferencing cross-validated against Edge Impulse cloud classifier; VoiceAge MFCC regression verified through train/holdout splits; mechanical fixtures inspected with feeler gauges and dial indicators for tolerance drift.
What are the implications?
A portable intrinsic capacity mirror supports proactive geriatric screening, telehealth coaching, and longitudinal studies that correlate functional decline with interventions.
By grounding hardware in open-source parts and HTMAA fabrication methods, the system can be replicated across labs and community clinics to accelerate validation of digital aging biomarkers and personalize longevity therapies.
Design
Your project should incorporate 2D and 3D design, multiple additive and subtractive fabrication processes, electronics design and production, embedded microcontroller design, interfacing, and programming, system integration and packaging.
2D Design
2D design work for the multimodal intrinsic capacity assessment system:
- Cardboard origami tiling for circular mirror frame, optimized for kerf learned during Week 1 laser characterization.
- Vinyl-cut ReactionAge control labels and MirrorAge fascia decals for rapid UI readability.
- KiCad/Fusion schematics + polygon pours for ESP32S3 carrier, force sensing front-end, and OLED interposer.
- 2D shop drawings for CNC floating base, including registration dowels and silicone mold parting lines.
- Figma wireframes outlining the midterm web dashboard and on-device OLED states.
Tools Used: Inkscape, Fusion 360, KiCad, Adobe Illustrator, Figma
3D Design
3D design work for device components and integration:
- 3D printed torsional spring grip housings with embedded brass inserts for load cell alignment.
- Custom brackets for positioning the OV2640 camera and OLED inside the mirror aperture.
- Ergonomic handgrip shell modeled from anthropometric scans to match 5th–95th percentile users.
- Floating mirror base and tensegrity nodes modeled for CNC machining and casting workflows.
- Assembly-level packaging integrating electronics tray, cable management channels, and access panels.
Tools Used: Fusion 360, FreeCAD, OpenSCAD, PrusaSlicer
Fabrication Processes Integration
Additive: 3D printing for custom components
Subtractive: Laser cutting, milling for precision parts
Electronics: PCB design and production
Programming: Embedded microcontroller development
Bill of Materials
Where possible, you should make rather than buy the parts of your project. Complete breakdown of materials, components, and sourcing information.
Bill of Materials
Complete list of materials and components:
Electronics Components
- Seeed XIAO ESP32S3 Sense + castellated carrier board, ATmega32 ReactionAge controller, USB-C power backplane.
- SparkFun Qwiic button, FlexiForce A201 sensor, Bosch BHI260 wearable IMU, on-board PDM microphone.
- OV2640 camera module, 128×64 SSD1306 OLED, 1.54″ IPS debug display, Neopixel status ring.
- Buck converters (AP34063), LiPo charger/boost (MCP73831 + TPS61291), BLE-enabled wearable node (nRF52840 Feather).
Mechanical Components
- PLA+, PETG, and Markforged Onyx filament for grips, sensor cradles, and cable guides.
- Cardboard tiles, 4 mm Baltic birch, acrylic light guides, and Delrin spacers for laser-cut structures.
- Heat-set inserts, M3/M4 stainless hardware, brass threaded rods, silicone gaskets, piano wire for tensegrity cables.
- Mold Star 30 silicone molds, Drystone cast ribs, Roto281 alloy ballast weights, mirrored glass panel.
Cost Breakdown
Running total $216.86 (Week 9). Electronics 54%, mechanical 28%, molding/casting 18%.
Detailed line items live in the weekly documentation tables (e.g., ReactionAge BOM CSV) and the midterm Airtable snapshot referenced in the lab-meeting deck.
Make vs Buy
Strategic decisions on fabrication vs. purchasing:
✅ Made Components
- 3D printed grip spring cartridges, wearable charging dock, and camera bezel.
- Laser-cut origami mirror shell, ReactionAge control fascia, and PCB mounting plates.
- Custom ESP32S3 carrier PCB, force-sensing daughterboard, and pogo-pin programming jig.
- Integrated sensor tower combining OLED, camera, microphone, and button into a single module.
🔸 Modified Components
- Adapted GRPR open-source grip meter geometry to fit torsional spring, swapping load cell for force sensor film.
- Re-housed SenseCraft XIAO ESP32S3 camera board into custom mirror-friendly enclosure.
- Customized SparkFun Qwiic button firmware for debounce-free ReactionAge measurements.
❌ Purchased Components
- Standard passives, headers, JST cables, LiPo cells, and regulators (Digi-Key, CBA stockroom).
- Wearable IMU/BLE module and FlexiForce sensor (SparkFun, Adafruit).
- Seeed XIAO ESP32S3 Sense dev kit, Edge Impulse model access, and spare OV2640 modules.
Fabrication Process
Week-by-week fabrication rolled forward subsystems toward the integrated mirror:
Week 1 laser-cut origami tiles for the circular bezel; Week 2 PCB milling + soldering for ReactionAge;
Week 3–4 torsional spring 3D prints and sanding jigs; Week 5 copper-clad milling and reflow of the ESP32S3 carrier;
Week 6 ShopBot machining of the floating base; Week 8 resin + FDM camera enclosure build; Week 9 wax machining, Mold Star casting, and Drystone ribs.
Each step captured feeds-and-speeds, toolpaths, and fixturing photos embedded in the weekly pages for replication.
Testing & Validation
Validation combined bench instrumentation and statistical analysis: oscilloscope timing to verify ReactionAge jitter, Edge Impulse confusion matrices for FaceTTD and VoiceAge,
Instron pull tests for the torsional grip cartridge, IR thermography while curing Mold Star molds, dial-indicator checks on CNC-machined bases, and adhesive shear testing on mirror mounts.
Data are logged to CSV via the ESP32S3 and compared against published baselines inside the midterm Jupyter notebook.
Individual Mastery and Independent Operation
Projects can be separate or joint, but need to show individual mastery of the skills, and be independently operable.
Individual Mastery
Demonstration of individual skills across all course units:
2D and 3D Design
- Modeled the tensegrity floating mirror, torsional grip shells, and camera bezel in Fusion 360 and Onshape.
- Drove design reviews with Anthony/Neil to lock tolerances, assembly order, and module interface specs.
Fabrication Processes
- Personally executed laser cutting, ShopBot machining, Bantam PCB milling, silicone casting, and Drystone pours.
- Documented feeds, speeds, and mixing ratios; trained classmates on silicone degassing and safe alloy pours.
Electronics and Programming
- Designed and routed the ESP32S3 carrier in KiCad/Fusion 360, assembled via reflow, and validated with multimeter/logic analyzer.
- Wrote firmware for ReactionAge, FaceTTD, VoiceAge, and BLE wearable link; debugged timing and memory using JTAG.
Independent Operation
Project operates independently without external dependencies:
✅ Standalone Functionality
- ESP32S3 carrier powers and orchestrates sensors with no tethered laptop.
- Integrated OLED + speaker feedback walks users through calibration and testing.
- Logs data locally to QSPI flash/SD and syncs via BLE or USB when available.
✅ User Independence
- OLED UI provides large-font prompts, countdowns, and IC score summaries.
- Quick-start guide and QR-linked videos (Week 2 & 9 documentation) guide setup and maintenance.
- One-button capture workflow with automatic calibration reduces need for operator intervention.
✅ Documentation
- Weekly HTMAA pages house schematics, code, CAM files, and process logs.
- Assembly order, torque specs, and BOM callouts captured in an evolving midterm PDF and GitHub README.
- Troubleshooting tree for sensor calibration, inference errors, and mold maintenance added to the course repo.
Course Presentation
Present your final project, weekly and group assignments, and documentation.
Final Project Presentation
Complete presentation of the multimodal intrinsic capacity assessment system:
- Live demo: capture selfie, voice clip, grip squeeze, and reaction test; display fused IC score.
- Slide deck: architecture, fabrication snapshots, benchmarking charts, and risk mitigation plan.
- Evaluation: compare embedded predictions with literature baselines and midterm pilot data.
Weekly Assignments
Integration of weekly work into final project:
- Weeks 0–1: concept boards, origami shell, kerf characterization; Week 2: ReactionAge electronics; Week 3–4: 3D grip + scanning; Week 5: PCB design; Week 6: floating base machining; Week 7–8: edge AI pipeline; Week 9: molds/casts.
- Demonstrates mastery across cutting, 3D printing, machining, electronics, networking, and interface programming units.
- Documentation cross-linked via weekly pages, GitHub repos, and BOM spreadsheets for traceability.
Group Assignments
Collaborative work and individual contributions:
- Embedded programming group: authored workflow trade-off analysis and repo organization that seeded ReactionAge firmware patterns.
- Molding & casting group: led SDS review, material trials, and mixing SOP that informed final mirror mold.
- Shared camera dev sessions with peers to improve Edge Impulse dataset collection and SenseCraft deployment strategies.
Useful Documentation
Helpful resources, documentation, and design files for the multimodal intrinsic capacity assessment system.
Open Source Grip Strength Meter
A cost-effective $50 grip strength measurement system that can be further optimized for our multimodal assessment platform.
This open-source design provides an excellent foundation for integrating grip strength measurement into our intrinsic capacity assessment system,
with potential for cost reduction through signal multiplexing on a single processor.
Available Resources:
- Complete design files and schematics
- Arduino-based firmware and code repository
- 3D printing files for device housing
- Assembly instructions and documentation
- Calibration procedures and testing protocols
- Integration examples for data collection systems
Complete Intrinsic Capacity Coverage Analysis
Comprehensive analysis of how different digital biomarkers cover the five domains of intrinsic capacity (IC) as defined by WHO.
Legend:
✅ Strong coverage |
🔸 Partial/indirect coverage |
❌ Not covered
Multimodal Intrinsic Capacity Pipeline
Detailed technical pipeline for processing multiple digital biomarkers to generate intrinsic capacity scores.
Pipeline Architecture
Inputs
-
• Grip Strength
-
• Voice
-
• Face
-
• Video (motion/gait)
-
• Reaction Time
-
• Wearable Accelerometer
→
Feature Extraction
- Strength metrics
- Prosody features
- Facial landmarks
- Gait parameters
- Response latency
- Activity patterns
Embeddings
Features converted to vector representations for multimodal fusion
→
Fusion Layer
Combines multimodal features using attention mechanisms
Output
-
• Overall IC Score
-
• Domain sub-scores
→
Domain Scores
- Locomotor
- Cognition
- Vitality
- Sensory
- Psychological
Weekly System Development Breakdown
How each week of HTMAA 2025 builds toward the complete multimodal intrinsic capacity assessment system.
Week 0: Project Ideation
Initial concept development and planning
- Project planning and documentation structure
- Research direction and concept sketches
Week 1: Precision Cutting
Laser and vinyl cutting techniques
- Device housing components via laser cutting
- Sensor mounting brackets and enclosures
- Vinyl cutting for device labeling and UI elements
Week 2: Embedded Programming
Electronics basics and microcontroller programming
- Microcontroller programming for data collection
- Basic sensor interface circuits
Week 3: 3D Scanning & Printing
3D technologies for device components
- 3D scanning for custom component design
- 3D printing for device housings
Week 4: Electronics Design
EDA and schematic design
- PCB design for grip strength measurement
- Sensor interface circuits and signal conditioning
- Power management and data storage systems
Week 5: Electronics Production
PCB fabrication and assembly
- PCB fabrication and debugging
- Component assembly and testing
Week 6: Computer-controlled Machining
CAM and precision milling
- Precision components via milling
- Custom mechanical parts
Week 7: Input Devices
Sensor integration for data collection
- Force sensors for grip strength measurement
- Microphones for voice analysis
- Camera systems for facial expression analysis
- Reaction time measurement circuits
Week 8: Output Devices
Actuators and system integration
- Display systems for real-time feedback
- Haptic feedback for user interaction
Week 9: Molding & Casting
Forming and resin techniques
- 3D printing and molding for custom components
- Silicone casting for device components
Week 10: Mechanical & Machine Design
System integration and mechanical design
- Mechanical design for ergonomic device housing
- System integration and calibration protocols
Week 11: Networking & Communications
Connectivity and communication protocols
- Bluetooth/Wi-Fi connectivity for data transmission
- Wearable accelerometer integration and data fusion
Week 12: Interface & Application Programming
UI development and application programming
- Mobile app development for user interface
- Cloud integration for data storage and analysis
- Machine learning pipeline for IC score calculation
Week 13: Wildcard & Final Integration
Final orders and complete system deployment
- Final testing, validation, and documentation
- System integration and deployment
Camera Subsystem
Advanced camera system implementation using XIAO ESP32S3 Sense with real-time image processing, EdgeAI integration, and interactive selfie capture functionality.
System Overview
The camera system successfully captures images, processes them into bitmaps, and displays them on an OLED screen using advanced Floyd-Steinberg dithering algorithms.
Video Demonstrations
Camera Bot Selfie
Demonstration of the interactive selfie capture system with touch controls
Picture of Picture
Meta demonstration showing the camera system capturing its own display
Code Implementation
60-Second Auto Capture System
Automated camera system that captures and displays images every 60 seconds using advanced image processing techniques.
Pseudocode Implementation:
1. Initialize camera with PSRAM frame buffers
2. Configure OLED display (128x64 pixels)
3. Set up 60-second capture interval timer
4. In main loop:
a. Check if 60 seconds have elapsed
b. Capture image from camera
c. Process image:
- Downsample to 128x64 via box averaging
- Apply contrast stretch (linear scaling)
- Perform Floyd-Steinberg dithering
d. Display processed bitmap on OLED
e. Release frame buffer
5. Repeat process
Touch-Controlled Capture System
Interactive camera system with touch controls allowing manual capture triggers in addition to automatic timing.
Pseudocode Implementation:
1. Initialize camera and OLED display
2. Set up touch pins (GPIO1 & GPIO2) with threshold detection
3. Configure 60-second auto-capture timer
4. In main loop:
a. Update touch sensor readings
b. Detect touch press events (justPressed)
c. Check for capture trigger:
- Touch press OR 60-second timer elapsed
d. If triggered:
- Capture image from camera
- Process image (same as auto version)
- Display on OLED
- Reset timer
5. Continue monitoring for next trigger
How It Was Done
The camera system implementation began with code from Charles Lu's electronics production weekly assignment,
which was based on the official XIAO tutorial.
Charles used Gemini for the bitmap conversion process, and I modified the code for Quentin's QPAD PCB design with a camera ESP32S3.
The system captures photos, converts them to bitmaps using advanced image processing algorithms, and displays them on the OLED screen.
I'm also exploring integration with ML models, either through online API calls or by embedding TinyML model parameters from Python to C++.
Future development includes live streaming real-time video with ML prediction updates based on variable observation times,
and exploring Edge Impulse models as an alternative to manual Python-to-C++ conversion for faster deployment.
EdgeAI/TinyML Integration
Edge AI can also be implemented using Edge Impulse models, which may be faster than manually converting Python models to C++.
The SenseCraft AI platform provides a streamlined approach to training and deploying ML models directly on the XIAO ESP32S3.
Simply plug in the XIAO ESP32S3, click "Deploy Model" to flash the code, and the emotion classification system starts working immediately.
Development Discussion: ChatGPT Session on Camera System Development
ReactionAge Module
Latency testing pipeline that drives the reaction-time biomarker using custom firmware, milled PCBs, and calibrated UX prompts.
Build Snapshot
- ATmega32U4 control board milled in Week 2 with debounced trigger buttons and RGB countdown prompts.
- Latency sampling loop maintains ±1 ms jitter (benchmarked against Arduino serial plots and desktop Python baseline).
- Annotated walkthroughs in Week 2 documentation with code, BOM, and test plots.
Midterm Demo Assets
- Firmware source and serial logging notebook bundled in ReactionAge latency demo.
- Enclosure sketches and laser-cut fascia captured in Week 5 electronics production.
Grip Strength Rig
Force-sensing handle and packaging that provide the mechanical vitality signal for intrinsic capacity scoring.
Hardware Status
- 3D printed torsional handle iterations from Week 3 tuned for 0–40 kg range using internal compliant ribs.
- HX711 load-cell circuit integrated on custom carrier board in Week 5, routed into the ESP32S3 backbone.
- Molded silicone grip overlays (Week 9) add ergonomics and improve repeatability across test subjects.
Next Steps
- Finalize calibration script comparing readings to reference dynamometer.
- Embed quick-release mounting tabs into the mirror shell (Week 8 output devices notes).
Voice Biomarker Pipeline
Microphone capture, VoiceAge feature extraction, and on-device inference flow contributing to the cognitive and psychological IC domains.
Implementation Highlights
- PDM microphone breakout characterized in Week 7 input devices with FFT sweeps and noise floor measurements.
- Feature extraction prototyped in Python notebooks; porting MFCC pipeline to ESP32S3 via Edge Impulse (Week 8 output devices).
- Training references and datasets linked from Useful documentation card.
Pending Work
- Deploy inference bundle to the SenseCraft board alongside the camera stack.
- Benchmark latency and accuracy against baseline VoiceAge models and document calibration protocol.
Week 0 - Introduction & Design
Project ideation and initial concept development for bioprinting rejuvenated tissue and aging biomarker devices.
Project planning
Concept sketches
Research direction
System Integration Plans: Establish the foundational architecture for multimodal data collection by designing the overall system framework that will integrate all six digital biomarkers (grip strength, voice, face, video, reaction time, wearable accelerometer) into a cohesive intrinsic capacity assessment platform.
Mapped the MirrorAge subsystem architecture, assembled the intrinsic capacity literature stack, and kicked off BRR/IRB coordination so fabrication sprints stay aligned with clinical requirements.
Week 1 - Principles & Cutting
Version control, laser cutting, and vinyl cutting techniques applied to final project components.
Laser cutting
Vinyl cutting
Version control
System Integration Plans: Fabricate precision-cut housing components and mounting brackets for all sensor modules (force sensors, microphones, cameras, reaction time circuits) using laser cutting, while creating vinyl-cut labels and UI elements for device identification and user guidance.
Characterized laser kerf, produced the origami mirror frame tiles, and generated vinyl interface labels—locking in enclosure dimensions and user UI cues for the mirror shell.
Week 2 - Embedded Programming
Electronics basics and embedded programming for the aging biomarker device components.
Microcontrollers
Programming
Electronics
System Integration Plans: Develop embedded programming protocols for real-time data collection from all six biomarker sensors, implementing initial signal processing algorithms and establishing the communication framework for multimodal data fusion.
Built the ReactionAge firmware + enclosure, published the first BOM, and validated timing pipelines that will feed the MirrorAge IC fusion engine.
Week 3 - 3D Scanning & Printing
3D scanning and printing techniques for bioprinting components and device housings.
3D scanning
3D printing
AI tools
System Integration Plans: Create custom 3D-printed components for camera mounting systems and facial recognition hardware, while developing 3D scanning protocols for ergonomic device design that accommodates all sensor modalities in a user-friendly form factor.
Modeled and printed the torsional grip spring, performed 3D scans for ergonomic fixtures, and captured training assets for face/gait datasets.
Week 4 - Electronics Design
EDA and schematic design for the aging biomarker device electronics.
EDA tools
Schematic design
Circuit design
System Integration Plans: Design comprehensive PCB schematics that integrate force sensor circuits for grip strength measurement, microphone preamplifiers for voice analysis, camera interfaces for facial recognition, and timing circuits for reaction time assessment into a unified electronics platform.
Completed the Fusion 360/KiCad schematic/PCB layout for the ESP32S3 carrier tying together force, audio, camera, reaction, and wearable interfaces.
Week 5 - Electronics Production
PCB fabrication, debugging, and assembly for the biomarker device.
PCB fabrication
Debugging
Assembly
System Integration Plans: Fabricate and assemble the integrated PCB containing all sensor interfaces, implementing power management systems for continuous operation and establishing data storage protocols for the multimodal biomarker data collection system.
Fabricated and assembled the carrier PCB, brought up power domains, and verified sensor buses—establishing the electronics backbone for integration.
Week 6 - Computer-controlled Machining
CAM and milling for precision components and device housings.
CAM
Milling
Precision machining
System Integration Plans: Machine precision mechanical components for the integrated device housing using computer-controlled milling, ensuring proper alignment and mounting for all sensor modules while maintaining ergonomic design for user comfort during multimodal data collection.
Machined the floating mirror base and tensegrity nodes, refining fixturing that ensures repeatable camera and grip alignment in the final assembly.
Week 7 - Input Devices
Sensors and embedded architectures for data collection in the biomarker device.
Sensors
Input devices
Data collection
System Integration Plans: Integrate all six input sensor systems (force sensors for grip strength, microphones for voice analysis, cameras for facial recognition and gait analysis, reaction time circuits, and wearable accelerometer) into the unified data collection platform with real-time processing capabilities.
Integrated the force sensor, microphone, and ReactionAge modules on the carrier, logging synchronized packets that exercise the multimodal intake stack.
Week 8 - Output Devices
Actuators and system integration for the biomarker device outputs.
Actuators
Output devices
System integration
System Integration Plans: Implement output devices including display systems for real-time intrinsic capacity feedback and haptic feedback mechanisms for user interaction, creating an intuitive interface for the multimodal biomarker assessment system.
Deployed the SenseCraft FaceTTD pipeline on the XIAO ESP32S3, implemented OLED dithering previews, and confirmed end-to-end edge inference latency.
Week 9 - Molding & Casting
Forming and resin techniques for bioprinting molds and device components.
Molding
Casting
Resins
System Integration Plans: Create custom molded components for the bioprinting aspects of the project and develop specialized casings for sensor protection, ensuring the device can withstand continuous use during multimodal data collection sessions.
Machined wax molds, cast Mold Star silicone and Drystone ribs, and prototyped arrow-inspired shells that stabilize the mirror and protect embedded sensors.
Week 10 - Mechanical & Machine Design
Kits and mechanical design for the bioprinting and biomarker device systems.
Mechanical design
Machine design
System integration
System Integration Plans: Complete the mechanical design integration of all system components, implementing calibration protocols for sensor alignment and developing the complete mechanical framework that houses all six digital biomarker measurement systems.
[Week 10 progress and contributions to final project placeholder]
Week 11 - Networking & Communications
BLE, Wi-Fi, and communication protocols for the biomarker device connectivity.
BLE
Wi-Fi
Communications
System Integration Plans: Implement wireless communication protocols (Bluetooth/Wi-Fi) for seamless data transmission from all six sensor modalities, enabling real-time data fusion and establishing connectivity for the wearable accelerometer integration into the multimodal assessment system.
[Week 11 progress and contributions to final project placeholder]
Week 12 - Interface & Application Programming
UI and application development for the biomarker device interface.
UI design
Applications
User interface
System Integration Plans: Develop the complete user interface and application programming for the multimodal system, implementing the machine learning pipeline for intrinsic capacity score calculation and creating cloud integration for comprehensive data storage and analysis of all biomarker measurements.
[Week 12 progress and contributions to final project placeholder]
Week 13 - Wildcard & Final Orders
Final orders and wildcard week activities for project completion.
Final orders
Wildcard activities
Project completion
System Integration Plans: Complete final system integration, testing, and validation of the complete multimodal intrinsic capacity assessment platform, ensuring all six digital biomarkers work cohesively to provide accurate WHO-defined intrinsic capacity scores across all five domains (locomotor, cognition, vitality, sensory, psychological).
[Week 13 progress and contributions to final project placeholder]
Design Files
[Links to CAD/SVG/assets placeholder.]
Reflections & Learnings
[Reflection notes placeholder.]
Contributions
Acknowledgements and contributions that made this project possible.
Gladyshev Lab and Collaborators
Special thanks to the Gladyshev Lab and collaborators for the fruitful discussions that led to this multimodal intrinsic capacity assessment idea,
which supplements my PhD research goals in aging and longevity. The conceptual framework for integrating multiple digital biomarkers
to assess intrinsic capacity domains emerged from collaborative research discussions on aging biomarkers and healthspan assessment.
Ethical AI Use
Transparent documentation of AI assistance used in this final project work, following course guidelines for ethical AI usage.
Cursor · Midterm Final Project Update
Cursor AI distilled Week 0–9 documentation and the Oct 31 lab meeting deck into midterm-ready narrative, cost, and validation content, replacing every placeholder in finalproject.html.
Transcript archived in markdown and HTML (generated via scripts/md_to_html_converter.py) for transparency.
AI-Assisted Intrinsic Capacity Research & Design
ChatGPT was used to research and develop the multimodal intrinsic capacity assessment framework, including the comprehensive coverage analysis table and technical pipeline design.
The AI assisted with structuring the WHO-defined intrinsic capacity domains, identifying appropriate digital biomarkers, and designing the fusion architecture for multimodal data processing.
AI-Assisted Final Project Development
Cursor AI assisted with developing the complete final project page structure, implementing the multimodal intrinsic capacity assessment framework, and creating comprehensive documentation.
The AI helped with HTML structure, responsive design, weekly system integration plans, and organizing the technical documentation for the complete biomarker assessment system.
AI-Assisted Final Project Presentation Structure
Cursor AI assisted with finalizing the project presentation structure to ensure full compliance with MIT Academy project presentation requirements.
The AI helped implement all required sections including answering questions, design documentation, bill of materials, individual mastery requirements,
course presentation structure, and spiral model development approach visualization.
AI-Assisted Camera System Development
ChatGPT was used to discuss camera system implementation strategies, image processing algorithms, and EdgeAI integration approaches.
The AI assisted with understanding Floyd-Steinberg dithering implementation, touch sensor integration, and exploring TinyML deployment options for the XIAO ESP32S3 platform.
AI-Assisted Project Highlights and Camera Subsystem Development
Cursor AI assisted with adding the project highlights section featuring camera system achievements and creating a comprehensive camera subsystem section with detailed code implementations, video demonstrations, and EdgeAI integration documentation.
The AI helped with HTML structure, responsive design, image processing explanations, and organizing the technical documentation for the complete camera system showcase.
This work is licensed under a
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License