Scope
Use models, don't train them
The AI engineer works above the API line. Tokens, sampling, and attention appear only as cost, latency, and quality knobs you actually turn in production.
In active development
Build AI products that actually ship.
The open-source AI Engineering handbook for software engineers. Prompts, context engineering, retrieval, agents, evals, and production operations — concise, opinionated, eval-first. No ML PhD required, no sign-up, free under CC BY-SA 4.0.
12parts
10capstones planned
100%free & open
The thesis
Eval-first, or vibes-driven.
The defining shift in AI engineering is from "the demo looked good" to "the eval suite passed." This handbook teaches evaluation before RAG and before agents — because you cannot improve what you cannot measure.
Format
Concise, one concept per chapter
Chapters you finish in eight minutes and apply immediately. Code and diagrams before prose. Every chapter ends with something you can do.
Proof · roadmap
Ten production-grade capstones
Coming as Part 12: search engines, support agents, coding agents, voice agents — each graded on one shared production rubric, several on public benchmarks.
The path
From first model call to production.
Twelve parts in a deliberate order: models, APIs, prompting, eval fundamentals, context engineering, retrieval, agents, deep evaluation and observability, security, production, product — with capstones and the interview loop next on the roadmap.
Evals before agents
You build your first 50-example eval set in Part 4 — before touching RAG or agents. Nothing in this book ships without a way to know whether it works.
Agents, the honest version
Workflows first, agents as escalation. Tool design, MCP and its security model, voice agents, browser and computer-use agents, and when multi-agent is genuinely worth 15× the tokens.
Production is the point
Prompt caching economics, cost engineering, eval-gated rollouts, incident response, and the product metrics that decide whether your AI feature survives.
Interview-ready
A research-backed interview framework: GenAI system design, the new AI-assisted coding rounds, and the eval-design interview almost nobody prepares for.
Start at chapter zero.
The book is being written in the open, part by part. Read what's live, watch the rest land.