Technique
Reranking
Second-stage re-scoring of retrieved candidates with a more expensive model for better top-k.
Definition
Reranking applies a cross-encoder or LLM scorer to the top 50–200 candidates returned by a cheap retriever, producing a tighter top-k. It dramatically improves RAG quality; Cohere Rerank and Voyage Rerank are popular hosted options.
Common use cases
- RAG quality
- Search precision
- Question answering