What problem does it solve? Choosing chunking strategies, embedding models, and vector databases by intuition leads to RAG systems with unmeasured retrieval quality. This Skill replaces guesswork with deterministic tools that analyze your actual corpus, generate a pipeline design from requirements, and evaluate retrieval with precision@k, recall@k, MRR, and NDCG metrics. ## Core Features & Use Cases - Corpus-Driven Chunking: Run chunking_optimizer.py on real documents to compare fixed-size, sentence, paragraph, and semantic strategies with size, boundary, and coherence metrics. - Requirements-Based Pipeline Design: Feed a requirements JSON into rag_pipeline_designer.py to get component recommendations, a Mermaid architecture diagram, config templates, and cost estimates flagged for verification. - Retrieval Evaluation: Score a TF-IDF baseline against ground truth with retrieval_evaluator.py, including failure analysis and improvement recommendations. - Use Case: Given a folder of technical documentation and a target of precision@5 ≥ 0.8, run the three tools in sequence to produce a validated RAG design with measured retrieval quality. ## Quick Start Analyze the documents in my docs folder, design a RAG pipeline for 5000 technical documents with high accuracy priority, and evaluate retrieval quality against my ground truth queries.
