# Hybrid search and reranking interview questions > 6 hybrid search and reranking interview questions for AI and ML engineer roles, each with a model answer. Canonical: https://llmdojo.dev/interview/topic/hybrid-and-reranking Last updated: 2026-10-23 6 questions: 1 junior, 3 mid-level and 2 senior. Taught in [Hybrid Search and Reranking](https://llmdojo.dev/learn/retrieval-knowledge-systems/hybrid-search-and-reranking). ## [What is hybrid search?](https://llmdojo.dev/interview/q/what-is-hybrid-search) Level: Junior Combining **term-based** and **embedding-based** retrieval, because they fail differently. **Diagram: Two ways to combine retrievers** 1. In sequence: a cheap, less precise retriever fetches a candidate set. AI Engineering's example: every document containing “transformer”. 2. A more precise, more expensive step orders the candidates: here, vector search picks the ones about the neural architecture. This second step is also called reranking. 3. The catch: anything the first step didn't fetch is gone for good. 4. In parallel: run several retrievers on the same query, each producing its own ranking. 5. Combine the rankings into one, for example with reciprocal rank fusion. A document only one retriever found can still make the final list. Either in **sequence** (a cheap retriever fetches candidates, a precise step orders them) or in **parallel** (both rank, then the rankings are fused). ## [How does reciprocal rank fusion work, and why use ranks rather than scores?](https://llmdojo.dev/interview/q/how-does-reciprocal-rank-fusion-work-and-why-use-ranks-rather-than) Level: Mid-level Each retriever gives a document **1 / (k + rank)**; sum across retrievers. BM25 points and cosine similarities aren't on the same scale, but ranks always are. **k = 60** keeps one retriever's first place from dominating, so agreement wins. Cormack et al. found k = 60 near-optimal but not critical. ## [When can fusion make retrieval worse?](https://llmdojo.dev/interview/q/when-can-fusion-make-retrieval-worse) Level: Mid-level When one retriever returns **noise** for a class of queries. On the course's six queries, BM25 matched a single common word (“every”, “phone”) in unrelated articles, those spurious matches got votes, and fusion ranked the right article first for **3 of 6**, against **5 of 6** for embeddings alone. Fusion is only as good as its inputs: measure it on your queries. ## [What's the difference between an embedding model and a reranker?](https://llmdojo.dev/interview/q/what-s-the-difference-between-an-embedding-model-and-a-reranker) Level: Mid-level An embedding model encodes query and document **separately**, so document vectors are precomputed and search scales to millions. **Diagram: Embedding model versus reranker** 1. An embedding model encodes the query and each document separately. 2. Relevance is a comparison between two vectors that never saw each other's text. 3. Because document vectors don't depend on the query, they're computed once. That's what makes searching a large collection fast. 4. A reranker reads the query and one candidate together, as a single input, so every word of one can attend to every word of the other. Hands-On LLMs calls this a cross-encoder. 5. It outputs a relevance score. Nothing can be precomputed, so it runs once per candidate at query time: affordable for a shortlist, not for the whole collection. A reranker is typically a **cross-encoder**: it reads query and candidate **together** and scores relevance. More precise, but one model run per candidate at query time, so it only sees a shortlist. ## [How do you size the candidate pool for a reranker?](https://llmdojo.dev/interview/q/how-do-you-size-the-candidate-pool-for-a-reranker) Level: Senior From where the right chunk usually lands. A reranker can only reorder what it's given: if relevant chunks typically sit at rank 8, a pool of 3 can't help. Each extra candidate is another model run, so the pool trades quality against latency, as Nogueira et al. note for multi-stage ranking. Measure recall at the pool size, then precision at the top after reranking. ## [A reranker fixes most of your misses but still prefers “safe for children over three” to “not safe for children under three”. What do you conclude?](https://llmdojo.dev/interview/q/a-reranker-fixes-most-of-your-misses-but-still-prefers-safe-for) Level: Senior A reranker is a better ranker, not a reasoner. On the course's data it ranked the right article first for **5 of 6** queries, but scored the reassuring toy article **4.66** against **3.89** for the one that answers. So keep more than one candidate, instruct the generator to read carefully and answer only from the passages, and keep negation cases in the evaluation set. ## Related - [All interview questions](https://llmdojo.dev/interview)