Design observability for retrieval, reranking, and answer grounding.

Instruction: Describe the traces and metrics you would expose for a production RAG stack.

Context: Assesses whether the candidate can design a practical architecture and explain the main tradeoffs. Describe the traces and metrics you would expose for a production RAG stack.

Official answer available

Preview the opening of the answer, then unlock the full walkthrough.

I would instrument the RAG pipeline as a sequence of observable decisions, not one opaque request. A useful trace should tell me what query we received, how it was rewritten, what candidate documents were retrieved, how they were reranked, which evidence actually entered the prompt, and what citations the final...

Upgrade to view official answer

Related Questions