Combines vector similarity and keyword-based search to improve retrieval accuracy and recall in RAG systems.
This skill provides standardized implementation patterns for hybrid search, a technique that merges the semantic understanding of vector embeddings with the precision of keyword matching (BM25). It includes ready-to-use templates for Reciprocal Rank Fusion (RRF), linear scoring, and integration with popular data stores like PostgreSQL (pgvector) and Elasticsearch. This is an essential toolkit for developers building Retrieval-Augmented Generation (RAG) pipelines who need to handle both broad semantic queries and specific technical terms, names, or codes that vector-only search might overlook.
Características Principales
01Implementation of Reciprocal Rank Fusion (RRF) and Linear Combination scoring
02PostgreSQL/pgvector integration for combined semantic and full-text search
03Elasticsearch patterns for dense vector and BM25 hybrid queries
04Cross-encoder reranking templates for high-precision results
05Normalization techniques for combining disparate search scores
060 GitHub stars
Casos de Uso
01Improving RAG system recall for technical documentation containing specific version numbers or error codes
02Building enterprise search engines that require both semantic matching and exact keyword hits
03Optimizing e-commerce search to balance product category similarity with exact SKU matches
