179 questions
No questions match those filters.
← All questions
EvaluationVery Hard· RAG Evaluation· Recall· Multi-hop Reasoning· System Debugging
Retrieval owns recall@5 of 0.94 and says the generator...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansThe current metrics are misleading. Recall@5 measures topical relevance to the surface query, which makes it silent on critical failure modes such as multi-hop reasoning errors or information staleness. Because the p95 budget of 800 ms prevents adding additional round-trips to the query process, the teams must stop relying on surface-level metrics. The resolution requires updating the evaluation set to specifically annotate evidence chains and timestamps to determine if the retrieval or the generation is truly at fault.