Combines vector similarity and keyword-based search to improve retrieval accuracy and recall in RAG systems.
This skill provides production-ready implementation patterns for merging semantic vector embeddings with traditional lexical keyword search (BM25). It is specifically designed to solve retrieval challenges in RAG systems, such as handling domain-specific jargon, technical codes, and names that pure vector search often misses. The skill includes optimized templates for Reciprocal Rank Fusion (RRF), linear score combination, and multi-stage retrieval using PostgreSQL with pgvector, Elasticsearch, and Cross-Encoder reranking to ensure the highest possible relevance for AI-driven applications.
主要功能
01Score normalization and fusion weighting strategies
02Reciprocal Rank Fusion (RRF) and Linear Combination algorithms
03Elasticsearch hybrid search and sub-search configurations
04Multi-stage retrieval patterns with Cross-Encoder reranking
050 GitHub stars
06PostgreSQL integration templates using pgvector and FTS
使用场景
01Implementing high-precision retrieval pipelines for agentic workflows
02Building enterprise search engines for domain-specific documentation
03Optimizing RAG systems where semantic search misses specific technical terms
