- Create a Conversation Model that front and backend can use
- I want to be able to do more than a one-off question answer
- The server needs to be able to respond with multiple messages for a given question
- This should allow polling on the frontend
- Maybe this is a Sqlite DB, that we keep in mem for now, but could persist later
- Add 'model' layer to be able to test different strategies
- Current model is one. ReAct model will be seperate
- Make UI look a bit more chat like
- Make sure to flush old convos from the in Mem DB
- Be able to enter a search and find relevant sentences via embeddings
- Feed question and context into ChatGPT for Answer
- Add Yew Frontend
- Add embeddings for full posts
- When we started I decided to do embeddings at the sentence level. But from reading more about LLM and embeddings it seems like I can/probably want to use bigger chunks