Design a rubric for judging grounded answers in a RAG assistant.

Instruction: Describe how you would build a grading rubric for a retrieval-based assistant.

Context: Assesses whether the candidate can design a practical architecture and explain the main tradeoffs. Describe how you would build a grading rubric for a retrieval-based assistant.

Official answer available

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

I would start the rubric with the core question: did the answer stay inside what the retrieved evidence can actually support. From there I would grade claim support, citation usefulness, completeness relative to the question, handling of uncertainty, and whether the answer narrowed or abstained...

Upgrade to view official answer

Related Questions