Rag And Retrieval easy
You are building a RAG assistant over a company wiki. A teammate proposes 4,000-token chunks so that each chunk 'has all the context.' What is the main downside of this choice?
- A Large chunks lower recall because HNSW indexes cannot store the longer vectors these chunks produce.
- B Retrieval becomes more precise, but rerankers can no longer score chunks longer than 800 tokens.
- C Each embedding averages many topics, diluting relevance and wasting context on unrelated text.
- D Large chunks only hurt lexical search; dense vectors represent long passages without any loss of meaning.
Answer: C. Each embedding averages many topics, diluting relevance and wasting context on unrelated text.
**Chunk size trades recall against precision and context cost**: one vector for 4,000 tokens blurs many topics together, and each retrieved chunk spends a large share of the prompt on irrelevant text. Option A is tempting, but vector length depends on the embedding dimension, not on chunk length. Larger chunks do make sense when answers need long, continuous context, such as legal clauses, and the corpus is small.
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?