Hello everyone,
I’m developing an assistant (initially with OpenAI) and I want to create this scenario:
- give him one json object through RAG and THEN I will provide the exact answer it has to give me
- it has to store in the memory
- so the next time I will text it, and I will upload new json object, it answers me in the exact way.
The problem I see that it seems you can use RAG or FEW-shots examples as a training way.
I want to understand if in the OpenAI env there is a way to mix both of them (because it isn’t working as I thought: in the chat.completions you can handle the few shots part, meanwhile where you create an assistant, you can upload files)
If you can help me out, I would really appreciate it