Rag And Retrieval easy
Your vector search retrieves the top 100 chunks, and the LLM receives the top 5. Evaluation shows the correct chunk is usually somewhere in the 100 but rarely in the 5. What should you add?
- A A larger first-stage k of 500 so the correct chunk shows up more reliably among the candidates.
- B A cross-encoder reranker that rescores the 100 candidates and keeps the best 5.
- C A second vector index with a different distance metric, taking the union of both top-5 lists.
- D Pass all 100 chunks to the LLM and instruct it to ignore anything that looks irrelevant.
Answer: B. A cross-encoder reranker that rescores the 100 candidates and keeps the best 5.
High recall at 100 with poor top-5 precision is the classic sign that a **reranker** is missing. A cross-encoder scores each query-chunk pair jointly and is far more precise than the vector distance, at a cost of tens to a few hundred milliseconds. Option A fixes a recall problem you do not have; if the right chunk were missing from the 100, widening k or adding hybrid search would be the right move.
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?