I am trying to build a RAG service (basically an assistant for internal documents within the company) and expose it as a backend service.
I tried using the assistant API with file knowledge it gives good results sometimes, but I was wondering what the best practices or better techniques are.
I am thinking of building a general solution where you can store the files (maybe in a vector) then give some instructions (as a system), then let the user enter a prompt.
The results should only be based on the files.
Any suggestion or concepts to explore?