A reranker takes a small set of candidate documents (say, top 50 from vector search) and scores each one against the original query using a cross-encoder model. Cross-encoders are slow but accurate — they read the query and the document together, while embedding models score them independently.
The pattern that works: retrieve top-50 fast with an embedding model, rerank to top-5 with a cross-encoder, send those five to the LLM. Reranking is usually the single biggest quality lever in a RAG system after chunking.
Bring this to your business
Knowing the term is one thing. Shipping it is another.
We do two-week AI Sprints — one term, one workflow, into production by Day 10.