Design a RAG question-answering agent over a firm's documents

Design a system that answers questions from a large, private collection of documents (for example a law firm's memos) with retrieval-augmented generation, citing the passages it used.

Walk through how documents are split into chunks and indexed, how a question is turned into a search, how the retrieved passages are put together into the model's context, and how the answer and its citations are produced. Then explain how you would evaluate it, measuring retrieval recall on its own and separately from the quality of the final answer, and how you keep the index current as documents change and respect who may read which document.