RAG & knowledge

Re-ranking

Re-ordering retrieved candidates with a more expensive model to put the best matches at the top.

also known as: reranker

In depth

First-pass retrieval (vector or hybrid) is fast but coarse. A re-ranker is a more expensive model that re-scores the top N candidates against the query, producing a better-ordered list. Cohere Rerank, Jina Reranker, and bge-reranker are the popular options. Worth the latency for production support bots, overkill for fast triage.

Related concepts

RAGA pattern where relevant documents are fetched from a knowledge base and pasted into context before the LLM answers.Hybrid searchCombining vector similarity with keyword matching to get the best of both.

Newsletter

Get the next post in your inbox.

Engineering notes from the Digitorn team. No marketing, no launch announcements, no "10 prompts that will change your life". Just the things we write that we'd want to read.

More in RAG & knowledge

Chunking/glossary/chunking Embedding/glossary/embedding Hybrid search/glossary/hybrid-search Knowledge base/glossary/knowledge-base RAG/glossary/rag Semantic search/glossary/semantic-search