Re-Ranking overview

Re-Ranking as a cross-cutting topic — applying cross-encoder models to re-score retrieved documents for relevance after initial retrieval. Covers the bi-encoder (fast, approximate) vs cross-encoder (slow, accurate) trade-off, popular re-ranker models (Cohere Rerank, BGE Reranker, cross-encoder/ms-marco), ColBERT-style late interaction for efficient re-ranking, and the retrieve-then-rerank pipeline pattern where a cheap first-stage retriever fetches candidates and an expensive re-ranker selects the best.