How to Build a Production-Ready RAG Pipeline With Modular Architecture

A retrieval-augmented generation (RAG) pipeline connects several services — document ingestion, text chunking, embedding generation, vector storage, and LLM-based response generation — into a unified system. A developer writing on DEV Community shared a production architecture where each component runs as a separate Docker container deployed via Cloud Run. The pipeline uses LangChain building blocks alongside Pinecone as the vector store, OpenAI embeddings, and FastAPI for orchestration, with all settings managed through an external YAML config file for easy component swapping. A key issue encountered was a SQLAlchemy session leak that silently exhausted database connections after thousands of requests, resolved by scoping sessions to the request lifecycle. The author emphasizes that modular design, hybrid search capability, and observability are critical for keeping such pipelines stable in production.