Final Project


👉 HTMAA Final Week Documentation


Final Idea: AI Pin

The key to really increasing the usability of AI for humans is context.

Understanding context is hard. We need to keep track of what is going on around and within a person.

Largely to that end, a couple of companies are currently releasing AI pins and related wearables.

Not sure what this can do for 699$ + 25$/m, what ChatGPT API can't do inside a app or some 5$ ESP microcontroller and a 1$ cam...
Cool idea but i don't get the pricing...

— David Krammer (@davidkrammer_) November 9, 2023

I just built the world’s most personal wearable AI!

You can talk to Tab about anything in your life. Our computers are now our creative partners! pic.twitter.com/RraiOPL4sm

— Avi (@AviSchiffmann) October 1, 2023

So many of the key hardware features of these pins are also present in the Xiao ESP32-E3 Sense. I feel challenged to make my own DIY cheap AI pin with it.



Inspiration: Star Trek Combadge

Much of my inspiration for the design still comes from my Idea #2, the Star Trek pin, which will be very similar (just with some AI capabilities).

The Combadge sells on Amazon for $65. Anyone I tell about this think that’s crazy. But it is done neatly - it’s very light and small. So I ordered it as a model.

Combadge Amazon Combadge arrived
Combadge example Combadge on me

It also has a really nice overview of the features:

Combadge overview Combadge parts

The way the Combadge works is that it has a microphone and a speaker. It listens for the “Hey, Siri” keyword and then sends the audio to the phone to be processed. The phone then sends the response back to the Combadge, which plays it through the speaker.



More inspiration: Walnut Speaker

There is amazing tiny walnut speaker by Penguin DIY.

How we go about this

Core:

Nice to haves:

Integrating the Weeks

My input and output weeks are related.

In the input week, I tested the camera and in the output week the speaker.

Camera Output week speaker

I realized I could just use the H-bridge board for my speaker, which lead to my first prototype.

Me soldering First prototype

In networking week, I connected my board to an external API via urequests. Networking API

Design

Pendant 1 Pendant 2

Iron Man Arc Reactor Heart

Pendant 3 Pendant 4

Previous Ideas

Dec. 13, 2023

HTMAA Final Week Planning

Documenting day by day Wed Thu Fri Sat Sun Mon 💀Tue💀 Spirals: Document every day TBD: Microphone pipeline Get audio from microphone –> used Seeed’s WAV Recorder example Stream audio to whisper –> saving to SD card and then uploading to API is fine for now (due to the small size of the audio files) Send transcription to GPT-4 Speaker pipeline Get response from GPT-4 Play audio to speaker Camera pipeline

Nov. 3, 2023

Idea 3: Give GPT a Face

Oct. 29, 2023

Idea 2: Star Trek Communicator Badge

Background Research: Bugging Devices and Spyware TCTEC Keychain Voice Recorder

Sep. 24, 2023

Idea 1: Flip Disc Display

When you walk into the Science and Engienering Complex (SEC) at Harvard, you will be greated by a magnificient display: The Harvard Time capsule. The time capsule is an interactive installation by the Art Group BREAKFAST, based on flip disc technology. It is attached to depth sensor cameras that allow the users to interact with the real-time changing display. The display “remembers” all past interactions and replays them - so no interaction is ever forgotten.