LLMs in Practice: Building RAG Assistants with Embeddings and Vector Databases: From Vector Search to Real-World AI Assistants

Build real-world AI assistants, not just toy demos.

LLMs in Practice shows you, step by step, how to build retrieval-augmented generation (RAG) systems with embeddings and vector databases, then turn them into production-ready assistants that search, reason, and take action.

Instead of hand-wavy theory, this book walks through a complete stack: ingesting documents, chunking and embedding them, storing vectors, wiring up retrieval, designing grounded prompts, evaluating quality, logging behaviour, securing data, adding tools, and finally deploying everything as a service. Along the way, you see the same patterns implemented in both Python and TypeScript, so you can work in whichever ecosystem you prefer.

You’ll learn how to take a messy folder of PDFs, wikis, and docs and turn it into:

• A searchable knowledge base backed by embeddings and a vector database

• A grounded RAG pipeline that cites its sources instead of hallucinating

• A tools-enabled assistant that not only answers questions, but can create tickets, trigger workflows, or call APIs

• An observable system with traces, logs, and a small evaluation set, so you can improve it over time

• A deployable service (FastAPI or Express) that real users can talk to

The focus throughout is on small, composable building blocks you can actually ship: tight retrieval functions, clear prompt templates, thin adapters around model providers, and simple web endpoints that wrap it all together. No heavy frameworks required.

By the end of the book, you’ll have a practical roadmap to go from “I can call an LLM API” to “I have a narrow, grounded assistant in production that my team actually uses”—and a set of patterns you can reuse for the next assistant you build.