What problem does it solve? Pure vector search misses exact keyword matches like names, codes, and domain-specific terms, while pure keyword search lacks semantic understanding. This Skill provides patterns and templates for combining both approaches to improve retrieval recall in RAG systems and search engines. ## Core Features & Use Cases - Fusion Methods: Implement Reciprocal Rank Fusion (RRF), linear score combination, cross-encoder reranking, and cascade filtering to merge vector and keyword results. - Database Templates: Ready-to-adapt implementations for PostgreSQL with pgvector and full-text search, and Elasticsearch with dense vectors and native RRF support. - Complete RAG Pipeline: A HybridRAGPipeline class that orchestrates parallel vector and keyword searches, fuses results, and optionally reranks with a cross-encoder. - Use Case: You are building a documentation search where users query exact error codes and natural language questions. Use the PostgreSQL template to run both searches in one SQL query with RRF fusion, then rerank the top candidates for final results. ## Quick Start Ask the AI to implement a hybrid search in PostgreSQL that combines pgvector similarity with full-text search using RRF fusion for your documents table.
