When should I use Reciprocal Rank Fusion (RRF)?

Use RRF when you need a robust, general-purpose method to merge results from different search algorithms (like BM25 and Vector Search) without having to manually tune weight parameters.

Does this skill support PostgreSQL implementations?

Yes, it includes comprehensive templates for implementing hybrid search using PostgreSQL with the pgvector extension combined with GIN-indexed full-text search.

What is hybrid search in the context of RAG?

Hybrid search is a technique that combines semantic vector search (which understands meaning) with traditional keyword search (which matches exact terms) to improve the accuracy and recall of document retrieval.

Why would I need a cross-encoder for search?

A cross-encoder acts as a reranker that evaluates the specific relationship between a query and a candidate document, providing much higher precision than initial retrieval methods alone.

Hybrid Search Implementation Patterns

Name: Hybrid Search Implementation Patterns
Author: gwickman

bygwickman

0•

データサイエンスとML

Implements high-performance hybrid search by combining vector similarity with keyword-based retrieval for RAG and search applications.

This skill provides standardized implementation patterns for hybrid search, effectively merging semantic vector embeddings with traditional keyword matching (BM25/FTS). It includes ready-to-use templates for Reciprocal Rank Fusion (RRF), linear scoring, and neural reranking using cross-encoders. Developers can leverage these patterns to build robust Retrieval-Augmented Generation (RAG) systems that handle both conceptual queries and specific domain-specific terminology across PostgreSQL and Elasticsearch environments, ensuring superior retrieval recall compared to single-method approaches.

主な機能

01Score normalization and fusion strategies for diverse search engines

02PostgreSQL/pgvector integration with full-text search (FTS) logic

03Cross-encoder reranking patterns for high-precision results

04Standardized Reciprocal Rank Fusion (RRF) and Linear Combination algorithms

05Elasticsearch hybrid search and RRF scoring configurations

060 GitHub stars

ユースケース

01Enhancing RAG systems with improved retrieval recall and precision

02Building domain-specific search engines that require exact keyword matching

03Implementing multi-stage retrieval pipelines with neural reranking

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add gwickman/auto-dev-test-target-1 hybrid-search-implementation

For use in Claude.ai and ChatGPT

Download Skill