Books I wrote with AI assistance.
Each one starts with a topic I researched in depth. The book is the consolidation: the ideas, patterns, and mistakes distilled into usefull lessons you can apply directly. They are shared here because the notes were more useful than I expected, and keeping them private helps nobody.
| Book | What it covers | Read it if |
|---|---|---|
| Building Reliable AI Applications | The ideas and operating model behind AI evaluation. Traces, failure taxonomies, datasets, evaluators, model-based judges, agents, regression gates, production loops, team structure, and a ninety-day rollout plan. | You want to understand why AI systems become reliable and how to organize the work. |
| AI Engineering Evals in Practice | The hands-on companion. Turn a product requirement into an eval contract, build a reproducible harness, add deterministic checks, calibrate an AI judge, evaluate retrieval and tool-using agents, gate releases in CI, and connect offline evals to production. Examples in Python with plain files. | You already know why evals matter and need to build them. |
| Internal Knowledge Bases in Practice | A hands-on guide from structured Markdown to a production, permission-aware MCP service. Covers content models, ingestion, compilation, provenance, freshness, hybrid and graph retrieval, authorization, context engineering, observability, evaluation, and operations. | You need to build a trustworthy company knowledge base that both people and AI agents can use. |
| Jev and Browser Agents in Practice | Typed decisions versus free text, the Jev decision model, the open-source browser-use and computer-use landscape, three integration patterns (browser automation, Slack agents with live search, runtime resolution), MCP tool contracts, prompt-injection security, benchmark literacy, and a rollout plan. | You are wiring a fast decision model into a browser-use or live-search agent and need to know what is real versus vendor claim. |
Each book is the output of a loop: research the topic, write a draft with an AI model, review every line against the sources and my own experience, rewrite, repeat. The AI accelerates the writing. The judgment about what is true and what is useful stays with me.
Errors are mine. If you find one, open an issue or a pull request.
Every book is a single Markdown file with a table of contents at the top. Read them on GitHub, in your editor, or convert them another format.
Free to read and share with attribution.