The craft of building with models.

The AI Engineering Handbook teaches software engineers how to build production-grade AI products: prompting, context engineering, retrieval, agents, evaluation, observability, security, cost, and the product skills that decide whether an AI feature survives contact with real users.

Who this is for

Software engineers who can build and ship a backend service and now want to build with foundation models, professionally. It is not an ML research curriculum: we do not teach how to train models, and internals appear only where they surface as cost, latency, or quality knobs you actually turn.

What makes it different

• Eval-first. You build your first eval set before you build your first RAG system. Everything after that iterates against measurement, not vibes.
• Concise. One concept per chapter, typically a few minutes to read, always ending with something you can do. We never pad.
• Opinionated. We name defaults and defend them. Decision rules, not feature matrices.
• Built for production. Every part ends in something deployable — traced, evaluated, budgeted, and secured. A capstone track (ten real builds on one shared rubric) is next on the roadmap.

The family

This is the third book under handbook.academy, alongside The HLD Handbook (system design) and The DSA Handbook (algorithms). Where a topic is architecture-scale, we cross-link HLD instead of repeating it; where it touches data structures, we cross-link DSA.

License and contributions

All content is free to read, with no sign-up, under CC BY-SA 4.0. Found an error or want to improve a chapter? Open an issue or a pull request on GitHub.