proskillv2.0.0✓ verified
rag-patterns
Deep, practical guide to retrieval-augmented generation that actually grounds answers — chunking, embeddings, vector + hybrid/BM25 search, reranking, context assembly, citation, and evaluation. Includes worked configs, a grounding prompt template, and a runnable chunker.
$npx vanara install rag-patterns
Included with Vanara Pro ($10/mo) — unlocks all 206 items. See pricing →
Overview
RAG grounds a model in *your* data so it answers from retrieved facts instead of parametric memory. The single most important truth in this skill: most RAG quality problems are retrieval problems, not generation problems. If the right chunk never enters the context window, no amount of prompt tuning will save the answer. This package is the deep reference — pipeline detail lives in references/, co
What it covers
- The pipeline (mental model)
- Chunking
- Embeddings & the vector index
- Retrieval: dense, sparse, hybrid
- Reranking
- Context assembly
- Grounded generation & citation
- Evaluation