Your question is RAG for High Retrieval Accuracy. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How would you architect a RAG pipeline to ensure high retrieval accuracy over domain-specific medical documents?
Explain how you would ingest and chunk documents, select embeddings and indexes, combine lexical and semantic retrieval, rerank results, and construct the generation prompt. Address medical terminology, document versions, metadata and access controls, hallucination, prompt injection in retrieved text, and refusal when evidence is insufficient. Define offline and online evaluation methods, then discuss latency and cost tradeoffs. Provide a practical architecture and implementation outline rather than only listing technologies.