What are you guys finding works well on the RAG Q&A side these days?
I made these changes and it seems to be effective:
-pass in last 4 Q&A pairs to similarity search
-summarize last 4 Q&A pairs + sources to add to system prompt as “Context of user’s conversation”
-pass in context as normal from the similarity search results

I'm getting this issue on a huge document, any ideas?
split_text_with_overlaps
raise ValueError(
ValueError: A single term is larger than the allowed chunk size.
Term size: 4105
Chunk size: 3811Effective chunk size: 3811