Design a benchmark harness for multi-hop and ambiguous RAG queries.

Instruction: Explain how you would build a benchmark that reflects the hard parts of retrieval-based assistants.

Context: Assesses whether the candidate can design a practical architecture and explain the main tradeoffs. Explain how you would build a benchmark that reflects the hard parts of retrieval-based assistants.

Updated

Official answer available

Read the opening below, then unlock the full answer and practical guidance.

I would build the harness around realistic failure structure, not just one gold answer per query...

Upgrade to view official answer

Related Questions