1000s of pages is the key here. We all know RAG can only take so far! So, I am looking into agentic search based retrieval using search tools for 1/ token efficiency to keep the costs low 2/ better context management (Double edged sword) I am looking for examples in BAML where people have done this and found a reliable pipeline mechanism.

The biggest challenge that I see is

passing the right context

to the model. And how the model is able to traverse the document in every retrieval step

and other ways to improve base rag