What is Reranker? A second-stage scoring model that re-orders retrieved chunks for relevance, typically a cross-encoder that scores each (query, chunk) pair more accurately than the initial vector search. Reranker explained Vector search (bi-encoder) is fast but imprecise. A reranker (cross-encoder like Cohere Rerank, BGE Rerank, or Voyage Rerank) takes the top-50 retrieved chunks and scores each against the query in a single forward pass. The reranked top-5 is what is actually given to the LLM. It adds some latency, but re-ranking can lift quality a lot: a BERT re-ranker beat the previous state of the art by 27% (relative) in MRR@10 on the MS MARCO passage ranking task (Nogueira and Cho, 2019). How Groovy Web uses Reranker We add a reranker when retrieval precision is not good enough in testing, and we benchmark the options (Cohere, BGE and others) on each client's own documents. Explore our RAG System Development service → Reranker: common questions What is Reranker? A second-stage scoring model that re-orders retrieved chunks for relevance, typically a cross-encoder that scores each (query, chunk) pair more accurately than the initial vector search. What should I know about Reranker? Vector search (bi-encoder) is fast but imprecise. A reranker (cross-encoder like Cohere Rerank, BGE Rerank, or Voyage Rerank) takes the top-50 retrieved chunks and scores each against the query in a single forward pass. How does Groovy Web use Reranker? We add a reranker when retrieval precision is not good enough in testing, and we benchmark the options (Cohere, BGE and others) on each client's own documents. Which terms are related to Reranker? Reranker is closely related to Retrieval-Augmented Generation (RAG), Vector Database, Semantic Search and Hybrid Search. Related terms Retrieval-Augmented Generation (RAG) Retrieval-Augmented Generation (RAG) is an AI architecture that combines a language model with a search system to answer questions using your own data. Vector Database A vector database is a specialized database designed to store, index, and search high-dimensional vector embeddings for AI applications like semantic search and RAG systems. Semantic Search A search technique that finds results based on meaning and intent rather than keyword matching, using embeddings to understand conceptual relationships between documents. Hybrid Search A search technique combining keyword matching with semantic similarity, leveraging both approaches to provide more relevant and comprehensive results than either alone. Knowledge Graph A knowledge graph is a structured representation of real-world entities and their relationships, used to organize information for AI reasoning and search. Agentic RAG RAG architecture where an AI agent decides when to retrieve, what to retrieve, and how to use the retrieved context, instead of retrieving on every query. ← More in Retrieval & RAG · All 107 glossary terms