Reranking

A module in a RAG pipeline that performs secondary sorting of candidate documents retrieved in an initial retrieval stage. It is typically implemented using a cross-encoder and achieves higher precision than vector similarity-based retrieval.