Building an AI assistant from scratch

An assistant that listens and uses your webcam to see what's around.

I built an AI assistant that hears and sees.

I used Python to build it. 169 lines of code and four main components:

• Gemini Flash is the LLM that I used. It’s much faster than GPT-4o, cheaper, and it’s concise in its answers. For some reason, GPT-4o is very chatty, and no amount of prompting helped me fix that.

• Whisper is the transcription model to turn aud…