rag-architect

Production-grade RAG system design covering chunking, embeddings, vector stores, hybrid search, reranking, and retrieval evaluation. For each step, validate before moving on (see checkpoints below). Load detailed guidance based on context: Checkpoint: assert all(c.metadata.get("source") for c in chunks), "Missing source metadata" Checkpoint: assert qdrant.count("knowledge_base").count == len(set(p.id for p in points)), "Deduplication failed" Checkpoint: assert len(hybrid_search("test query", tenant_id="demo")) > 0, "Hybrid search returned no results" Checkpoint: Target context_precision >= 0.7 and context_recall >= 0.6 before moving to LLM integration. When designing RAG architecture, deliver: Documentation - Guides five core workflow steps: requirements analysis, vector store design, chunking strategy, retrieval pipeline configuration, and quality evaluation with checkpoints - Supports multiple vector databases (Pinecone, Weaviate, Chroma, pgvector, Qdrant) with schema design, indexing, and sharding strategies - Implements hybrid search combining dense vector retrieval with BM25 keyword search, plus reranking via Cohere for top-k result refinement - Includes evaluation framework using RAGAS metrics (context precision, recall, faithfulness, answer relevancy) to validate retrieval quality before LLM integration - Provides reference guides for embedding model selection, semantic...