rag-retrieval-patterns

Configure dense, sparse, and hybrid document retrieval for RAG pipelines.

7|1|Updated Feb 14, 2026

Search Tags:#embeddings #rag #retrieval #grounding #reranking #vector-search #bm25

Ready to run

npx skills add https://github.com/fratilanico/apex-os-bad-boy --skill rag-retrieval-patterns

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: rag-retrieval-patterns
Source: https://github.com/fratilanico/apex-os-bad-boy/tree/main/rag-retrieval-patterns
Command: npx skills add https://github.com/fratilanico/apex-os-bad-boy --skill rag-retrieval-patterns

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps teams design, debug, and optimize retrieval-augmented generation (RAG) pipelines by selecting and combining dense, sparse, and hybrid retrieval strategies to improve excerpt relevance and grounding in documents.

Core Features & Use Cases

Decision-tree guidance for choosing between BM25, dense embeddings, and hybrid pipelines.
Concrete implementation patterns, including dense retrieval, BM25, and cross-encoder reranking.
Use cases include QA systems, document-grounded assistants, and knowledge-base search pipelines.

Quick Start

Describe a retriever setup and this skill configures a hybrid RAG pipeline leveraging dense, sparse, and reranking for your documents.

Frequently Asked Questions about rag-retrieval-patterns

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I choose between BM25 and dense embeddings for my RAG pipeline?▼

Choosing between BM25 and dense embeddings for a RAG pipeline depends on your query type: BM25 handles exact keyword matches while dense embeddings capture semantic meaning. This Skill provides decision-tree guidance to select the right retrieval strategy.

What is the best way to combine sparse and dense retrieval for document search?▼

The best way to combine sparse and dense retrieval for document search is a hybrid RAG pipeline. This Skill configures hybrid pipelines that merge BM25 and dense vector search, then applies cross-encoder reranking to maximize excerpt relevance.

How do I integrate vector search and reranking into a knowledge-grounded assistant?▼

To integrate vector search and reranking into a knowledge-grounded assistant, you describe a retriever setup and this Skill configures the hybrid pipeline. It supports modular components like embeddings models and cross-encoders to generate grounded answers with source cites.

When should I use cross-encoder reranking in a retrieval-augmented generation workflow?▼

Use cross-encoder reranking in a retrieval-augmented generation workflow when initial dense or sparse retrieval yields too many low-relevance documents. Reranking filters and reorders excerpts to improve grounding and answer accuracy for QA systems.

Why does my hybrid RAG pipeline return irrelevant excerpts despite using dense embeddings?▼

Irrelevant excerpts in a hybrid RAG pipeline often occur when dense embeddings alone miss exact keyword matches or lack reranking. Adding BM25 for sparse retrieval and cross-encoder reranking improves excerpt relevance and document grounding.