Hands-On RAG for Production: Design, Develop, and Deploy Production-Ready RAG Applications

A RAG prototype can look convincing long before it is ready for the demands of production. Hands-On RAG for Production focuses on the engineering decisions in between: how to prepare and retrieve useful information, evaluate the answers a system produces, and address the operational realities of deploying it at scale.

Ofer Mendelevitch and Forrest Sheng Bao take readers from the basic RAG stack to production architecture and newer approaches including agentic RAG, multimodal retrieval, and GraphRAG. The emphasis is not just on assembling components, but on understanding the trade-offs behind them.

Follow the RAG pipeline from data to answer

The early chapters explain the flow through a RAG application, then examine the pieces that shape its performance: document parsing, text chunking, embedding models, vector databases, vector search, and prompts for generative language models. Examples connect the concepts to practical tools and implementations, including LangChain, Sentence Transformers, pgvector, and Anthropic Claude.

Move beyond the proof of concept

Production systems face problems a quick demo can hide. The book considers document scale and quality, index freshness, cost, hybrid search, reranking, latency, data privacy, vendor integration, and the organizational work involved in a successful deployment. Its build-versus-platform discussion offers a framework for weighing DIY infrastructure against RAG-as-a-service options.

Evaluate what the system actually does

Retrieval and generation can fail in different ways, so evaluation needs to look beyond whether an answer sounds plausible. Dedicated coverage explores retrieval and generation metrics, LLM-as-a-judge approaches, human feedback, bias and safety, latency, uptime, and resource efficiency. These topics help teams make quality and reliability observable rather than relying on a promising first impression.

Explore agents, multimodal RAG, and knowledge graphs

Later chapters extend the discussion to AI agents, multimodal data, and knowledge-enhanced retrieval. Topics include agent workflows and tool calling, tables and images in documents, audio and video, and the use of knowledge graphs and GraphRAG. Together, these sections show how RAG designs can expand—and what new evaluation, accuracy, privacy, and cost considerations come with that expansion.

For engineers building beyond the demo

Classified by its publisher as intermediate to advanced, this book is aimed at engineers and architects who need to design or improve RAG applications in practical settings. It will be especially relevant to readers comparing system components, diagnosing retrieval or generation problems, and planning for production constraints.

For a team asking not only “Can we build a RAG application?” but “Can we make it dependable, measurable, and maintainable?”, this book provides a detailed engineering-focused path to consider.