Does this skill work with existing vector databases?

Yes, it includes native implementation patterns for popular platforms like Pinecone and Weaviate, as well as generic RRF functions that work with any retrieval source.

What is the role of the Reranker in this skill?

The reranker uses a Cross-Encoder model to evaluate the specific relevance of the query against the top documents returned by the hybrid search, significantly increasing precision for the final results.

How do I determine the right weights for dense and sparse search?

The skill provides a weight tuning guide: use higher sparse weights (0.7) for code and technical data, and higher dense weights (0.7) for conversational or general text.

When should I use hybrid retrieval instead of just vector search?

Use hybrid retrieval when your queries contain specific technical terms, unique identifiers, or error codes that semantic embeddings might overlook in favor of broader conceptual matches.

Hybrid RAG Retrieval

Name: Hybrid RAG Retrieval
Author: latestaiagents

bylatestaiagents

•

数据科学与机器学习

Implements high-precision search by combining semantic dense vectors with keyword-based sparse retrieval for optimized RAG results.

Hybrid RAG Retrieval is a specialized skill designed to solve the common 'missing match' problem in standard vector search. While embeddings excel at semantic context, they often fail to capture exact keywords, error codes, or domain-specific terminology. This skill provides standardized implementation patterns for merging BM25 (sparse) and Vector (dense) search using Reciprocal Rank Fusion (RRF) and Cross-Encoder reranking. It is an essential toolkit for developers building production-grade AI agents that require pinpoint accuracy across technical documentation, enterprise knowledge bases, and complex datasets.

主要功能

01Multi-provider patterns for LangChain, Pinecone, and Weaviate

02Configurable weighting guidelines for technical, legal, and conversational data

03Built-in retrieval evaluation metrics including MRR and Recall@5

042 GitHub stars

05Reciprocal Rank Fusion (RRF) for balancing multiple retrieval sources

06Cross-Encoder reranking logic for enhanced top-K precision

使用场景

01Legal and medical search engines where specific terminology is critical

02Technical support bots requiring exact error code and ID matching

03Enterprise knowledge bases that mix natural language queries with product nomenclature

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add latestaiagents/agent-skills hybrid-retrieval

For use in Claude.ai and ChatGPT

Download Skill