Reranking

What Reranking is

Vector search is fast but approximate. A cross-encoder reranker reads the query and each candidate together, producing a much more accurate relevance judgement on a small shortlist.

How it works

Retrieve a few dozen candidates cheaply, rerank them with a cross-encoder, and keep the top few for the prompt. Hybrid pipelines merge keyword and vector candidates before reranking.

Why it matters

It is often the highest return-on-effort upgrade in a RAG system, because it fixes the case where the right passage was retrieved but ranked too low to be included.

Common uses

• →Enterprise search
• →RAG precision improvement
• →Recommendation shortlists