Interview Guides
You are building an LLM-powered search experience for a mid-sized product team. The first pass retrieval is already good enough for many queries, but some answers still miss the best source or pick a weak passage. Your team is considering a re-ranking step before generation.
When is it worth adding the extra latency of re-ranking, and how would you decide?