What problem does it solve? Building RAG systems that return relevant context instead of hallucinations is hard: poor chunking, wrong embedding choices, and missing reranking silently degrade answer quality. This Skill provides expert guidance on designing retrieval pipelines where retrieval quality is measured and optimized separately from generation. ## Core Features & Use Cases - Semantic Chunking Strategies: Chunk documents by meaning and structure rather than fixed token counts, preserving sentence boundaries and context continuity. - Hierarchical & Hybrid Retrieval: Combine multi-level indexing (paragraph, section, document) with BM25/vector hybrid search and Reciprocal Rank Fusion for better precision. - Retrieval Evaluation & Anti-Patterns: Identify common failure modes such as embedding everything, skipping reranking, and never refreshing stale embeddings. - Use Case: When building a document Q&A feature over a company knowledge base, use this Skill to design the chunking strategy, select embedding models per content type, add metadata pre-filtering, and set up retrieval quality metrics before wiring the LLM prompt. ## Quick Start Ask the assistant to design a RAG pipeline for your document collection, including chunking strategy, embedding model selection, and hybrid search with reranking.
