Overall I’m happy with how this turned out! It’s very fast and it doesn’t hesitate to roast me when given the chance. It misunderstood “temps” as “tents” but this only really happens when I shorten words and don’t speak toward the microphone. Ideally it would pick this up so for the V2 Tile I might use three microphones instead of one. I’m using a lexicon to decide which emotions to display but I might switch to an AI model for better accuracy. My implementation of the emotions was a little half baked but that is ok because I’ll need to redo all that code anyway. I’m switching to WS2812 LEDs which work differently so hopefully I’ll be able to implement a more dynamic approach for displaying emotions. The server code is only 400 lines right now which is good because I want to keep it as simple as possible.

For the server code I’ll eventually want to implement a vector database for storing RAG context. This could allow 45 to retain context for months or years! I could add a web portal so I could upload documents to ask 45 questions about. All the big companies are using RAG so this is the best route.