Rag And Retrieval easy
Logs for a failing question show that the correct chunk was ranked in the top 3 retrieved results, yet the final answer was still wrong. Where should you focus first?
- A Switch embedding models, since a stronger embedder would have ranked the correct chunk first.
- B Increase top-k so that the model sees more supporting chunks alongside the correct one.
- C The prompt and generation step, since retrieval did its job and the model misused it.
- D Re-chunk the corpus with smaller chunks so the correct passage is more tightly scoped.
Answer: C. The prompt and generation step, since retrieval did its job and the model misused it.
**Evaluate retrieval and generation separately**: if the right evidence was in context, the failure is in prompting or generation, such as unclear instructions or conflicting chunks. Changing retrieval (options A, B, D) adds cost and can add noise without fixing the real issue. Retrieval changes would be the right focus only if the correct chunk were missing from the retrieved set.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- You must index 100M chunks using a 3,072-dimension embedding in fp32, served from RAM via HNSW. The raw vectors alone take about 1.2 TB before graph overhead. You need to cut memory several-fold while losing at most a point or two of recall@10. Which approach fits?
- A multi-tenant SaaS app stores all tenants in one vector index with a tenant_id field. Queries fetch the top 20 chunks by similarity and then drop chunks from other tenants. Small tenants often get 0-2 results. Why does this happen, and what is the fix?
- Runbooks are split into fixed 300-token chunks with no overlap. When a procedure step starts near the end of one chunk and finishes in the next, answers come back incomplete. What is the lowest-cost fix?
- Your offline eval shows recall@5 of 92% on a labeled set, so the gold passage is in the context for nearly every question. Yet human graders mark 35% of answers wrong, and the wrong answers often ignore or contradict the gold passage. Where should the team invest next?
- Your product requires every RAG answer to include clickable source citations that users can verify. How should those citations be produced?
- A runbook RAG system uses fixed 500-token chunks with no overlap. Users who ask for a multi-step recovery procedure get answers that stop at step 4 of 9, and retrieval logs show the top hit ends mid-procedure. What is the most targeted fix?