RAG & Retrieval · Ground answers in evidence
How does a reranker improve retrieval quality?
Short answer
A reranker scores query and passage jointly, which is more accurate than comparing independent embeddings. It reorders a bounded candidate set, improving precision at the cost of latency.
Bi-encoder retrieval embeds the query and document separately, so it never sees them together. A cross-encoder reranker reads both and produces a relevance score. That gives much better discrimination among candidates, but it cannot run over millions of documents in real time.
- Retrieve a broad candidate set for recall, then rerank for precision.
- Choose the candidate count based on latency budget and observed recall.
- Measure context precision and answer quality with and without reranking.
- Have a fallback if the reranker is slow or unavailable.
Follow-ups you should expect
Interviewers use these to check whether the first answer was understanding or memorisation.
- 1How many candidates would you rerank in a chat product?
- 2When would a reranker not be worth the latency?
- 3How would you evaluate the reranker's contribution?
Say it in the interview
A reranker reads the query and each candidate together, so it can judge relevance much more accurately than separate embeddings. I use it after a recall-oriented retrieval step, over a bounded set, to improve precision before packing the prompt. It adds latency, so I measure whether it actually improves answer quality. And I remember it cannot fix a document that retrieval never returned.
Use this as a model, not a script. Saying it in your own words is what makes it survive follow-up pressure.
Related questions
Reading is step one
Now say it out loud.
Understanding an answer and delivering it under pressure are different skills. Rehearse this topic in a live RAG & Retrieval mock interview and get scored feedback on depth, clarity and confidence.