Combines vector similarity and keyword-based search to optimize retrieval accuracy and recall in RAG systems and search engines.
This skill provides standardized patterns and implementation templates for hybrid search, bridging the gap between semantic understanding and exact keyword matching. It includes robust configurations for Reciprocal Rank Fusion (RRF), linear scoring, and advanced re-ranking using cross-encoders. Designed for developers building LLM applications, it offers ready-to-use code for PostgreSQL (pgvector), Elasticsearch, and Python-based fusion logic, ensuring high-quality document retrieval even when domain-specific terminology or precise identifiers are required.
主な機能
013 GitHub stars
02Elasticsearch hybrid search templates
03PostgreSQL pgvector and full-text search integration
04Reciprocal Rank Fusion (RRF) implementation
05Score normalization and linear weighting methods
06Cross-encoder re-ranking patterns
ユースケース
01Enhancing AI agent context windows with more accurate document selection
02Building search engines that combine semantic meaning with keyword relevance
03Improving RAG retrieval for technical documentation containing specific jargon or codes