Проектирование RAG-систем с векторным поиском

ai-agents

Rag architect is a Claude Code skill that designs and implements production-grade RAG systems — chunking documents, generating embeddings, configuring vector stores, building hybrid search pipelines, applying reranking, and evaluating retrieval quality end-to-end. Reach for it when a corporate knowledge base needs to feed an LLM without hallucinations, when semantic search returns irrelevant results and you need to diagnose why, when you must choose between Pinecone, Weaviate, Chroma, pgvector, and Qdrant for your scale, or when an existing pipeline has unacceptable precision and recall. The skill walks through five checkpointed stages: requirements and latency analysis, vector store schema design, chunking strategy selection, hybrid Vector + BM25 retrieval with Reciprocal Rank Fusion, Cohere reranking, and evaluation via RAGAS metrics (context_precision, context_recall, faithfulness, answer_relevancy). All implementation examples are in Python and draw on LangChain, Qdrant, OpenAI Embeddings, rank_bm25, and Cohere; related skills include python-pro, database-optimizer, and api-designer. The direction is building and tuning retrieval pipelines for AI applications — not training models or replacing classic full-text search where no AI layer is needed.