AI Engineering from Scratch
A free, open-source curriculum of 503 lessons across 20 phases, teaching you to build AI end-to-end from math to agents.
- Source repo
- rohitg00/ai-engineering-from-scratch
- Stars
- ★ 71k
- Last updated
- today
- License
- MIT
- Primary language
- Python
- FA score
- 33/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free, no paid service needed
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer who wants to understand AI algorithms from the ground up, implementing models instead of just calling APIs.
- Not a fit if
- Developers who only want to call APIs, not build algorithms from scratch
- Learners needing complete non-English lessons; only landing pages are translated
- Anyone wanting a quick reference rather than a ~340-hour curriculum
- Source review
- 33/100 · Major gaps 8 safety controls not found
What does this agent do, and when should you use it?
This repository is a comprehensive, free, open-source AI engineering curriculum organized into 20 phases and 503 lessons, covering Python, TypeScript, Rust, and Julia. It spans from math foundations and classical machine learning, through deep learning, computer vision, NLP, transformers, generative AI, reinforcement learning, LLM engineering, multimodal models, agent engineering, autonomous systems, multi-agent and swarms, to production infrastructure and safety. Each lesson follows a 'Build It / Use It' pattern: implement algorithms from raw math, then replicate using production frameworks like PyTorch, and ship a reusable artifact (prompt, skill, agent, or MCP server). The course is accessible via a website, as a cloned repository, or through agent skills that turn your coding agent into a tutor. A six-volume EPUB/PDF book series is also built via CI and attached to releases.
The repository provides structured course content: lesson narratives under phases/<num>-<phase>/<num>-<lesson>/docs/en.md, runnable code in code/, and artifacts like prompts, skills, agents, and MCP servers in outputs/. It integrates with any agent that reads SKILL.md via a skills/ directory, offering commands like /start-learning, /learn, /course-guide, /find-your-level, and /check-understanding to guide learners through placement quizzes, personalized plans, interactive lessons, and quizzes. Installation is via npx skills add rohitg00/ai-engineering-from-scratch or cloning the repo. The course can be consumed from the command line, the website (aiengineeringfromscratch.com), or by running lesson code directly (e.g., python phases/01-math-foundations/01-linear-algebra-intuition/code/vectors.py). The book can be built locally with python3 scripts/build_book.py.
- A developer who wants to understand AI algorithms from the ground up, implementing models instead of just calling APIs.
- A professional needing to use AI tools in their work but lacking systematic training and seeking a hands-on, curriculum-based path.
- A software engineer who wants to train their coding agent as an AI tutor, using interactive lessons and personalized study plans.
- A learner who wants to build a portfolio of reusable prompts, skills, agents, and MCP servers as they learn.
- A reader who prefers consuming the course as an EPUB/PDF book series for offline study.
How do you install or deploy this agent?
You can start learning without cloning: in any agent that supports SKILL.md (e.g., Claude Code, Cursor, Codex, OpenClaw, Hermes), run npx skills add rohitg00/ai-engineering-from-scratch and then use /start-learning. To clone and run locally, use git clone https://github.com/rohitg00/ai-engineering-from-scratch.git and install necessary dependencies; Python 3 is required for running lesson code, and Node.js is needed for npx. Detailed setup steps are in the README.
How do you use this agent?
In your agent, run /start-learning to take a ten-question placement quiz, which generates a personalized study plan saved to LEARNING.md. Then use /learn to go through lessons interactively (concept → math → code → quiz), and /course-guide <topic> to jump to specific lessons. Alternatively, read lessons on the website or clone the repo and run code, e.g., python phases/01-math-foundations/01-linear-algebra-intuition/code/vectors.py. Install all skills with python3 scripts/install_skills.py <target>.
What are this agent's strengths and limitations?
- Builds every algorithm from raw math, ensuring deep understanding
- Each lesson ships a reusable artifact (prompt, skill, agent, MCP server), building a portfolio
- Extremely broad coverage: from fundamentals to multi-agent and production, with 20 phases and 503 lessons
- Free forever with MIT license, and multi-language support in landing pages
- Agent integration provides an interactive, personalized learning experience directly in your coding environment
- Large time commitment (~320 hours) may feel overwhelming for beginners
- Requires setup of Python, Node.js, and optionally Docker, which may be a barrier for some
- English is the only complete language; other languages only translate landing pages, not lessons
- Being a curriculum, it focuses on learning rather than production-ready code, so you'll need to apply it yourself
How does this agent compare with similar options?
This curriculum positions itself as a comprehensive alternative to scattered tutorials and papers, focusing on building foundations rather than just using frameworks. It may be compared with other AI courses, but it stands out for its depth, breadth, and artifact-driven approach.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| AI Engineering from Scratch This agent | 33 · Major gaps | CLIFree | ★ 71k | today | Python | Codex · Claude Code |
| Swiftide | 68 · Some gaps | Library / SDKFree + model costs | ★ 793 | 1d ago | Rust | OpenAI API · Claude API |
| Open Interpreter | 45 · Major gaps | CLIFree + model costs | ★ 69k | 8d ago | Rust | Codex |
| Stream Vision Agents | 43 · Major gaps | Library / SDKFree + model costs | ★ 8.2k | 25d ago | Python | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows the repository is an educational curriculum, not an executable agent, so there are no mechanisms for least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution. All trust criteria score 0 because no trust controls are implemented.
Self-consistency: The curriculum structure is consistent, and README and test files show consistent naming and structure, but not all 503 lessons are verified, so score 2. Dependency availability: requirements.txt lists common dependencies but lacks locked versions or integrity checks, score 1. Failure messages: Test files have assertions but no user-friendly error messages, score 1.
Audience and scenarios: README clearly targets learners and provides multiple usage modes, score 2. Capability boundaries: No explicit statement of course coverage or limitations, score 1. Trigger precision: Skill commands like /learn are described but not detailed trigger conditions, score 1. Environment fit: Supports multiple agents and languages but lacks detailed environment configuration, score 2.
Information architecture: Clear directory structure with phase and lesson lists, score 2. Install notes: README provides installation commands, score 2. Naming stability: Consistent lesson naming but no version stability statement, score 1. Examples and FAQ: Example code present but no FAQ, score 2. Known limitations: Not mentioned, score 0. License: MIT license clear, score 2. Versioning and changelog: ROADMAP.md exists but no CHANGELOG, score 1. Maintenance responsibility: No explicit maintainer, score 1.
Output usability: Lessons produce reusable artifacts, score 2. Marginal value: Provides unique learning path, score 2. Cost-benefit: Free and open source but no cost analysis, score 1.
Claim traceability: Claims in README lack evidence, score 1. Cross-source corroboration: No external verification, score 1. Fact-inference separation: Not explicitly distinguished, score 1.
- Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
- Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
- This repository is an educational curriculum, not an executable agent product, so trust-related criteria are not applicable.
- Dependencies are not version-locked, posing supply chain risks.
- No known limitations or FAQ provided, users may encounter undocumented issues.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v2026.10.3 Edition 2026.10.3
Release notes excerpt
Edition 2026.10.3 adds a reverse engineering track, a third manual, and LaTeX math in the lessons. The curriculum grows from 523 to 542 lessons across 20 phases. Learn reverse engineering Fourteen new Phase 13 lessons teach how to read a compiled program you built yourself, from file bytes to evidence-backed pseudocode (#597). …View on GitHub ↗ -
v2026.10.2 Edition 2026.10.2
Release notes excerpt
Edition 2026.10.2 adds a second manual, a public API and MCP server for the curriculum, and fixes across the site. The curriculum stays at 523 lessons across 20 phases, with 2 lesson texts revised since Edition 2026.10.1. Read Docker Sandboxes and Docker Agent 101 [Docker Sandboxes and Docker Agent 101](https://aiengineeringfromscratch.com/manual-docker-sandboxes-101) explains Docker Sandboxes (sbx 0.47.0) and Docker Agent (docker-agent 1.149.0), from their purpose to each command, file, and policy object. It has 33 sections in 7 parts, 8 reference sections, and 34 figures. The PD …View on GitHub ↗ -
v2026.10.1 Edition 2026.10.1
Release notes excerpt
Edition 2026.10.1 adds the first long-form manual, a catalog of 100 projects, and fixes across the site. The curriculum stays at 523 lessons across 20 phases, with 3 lesson texts revised since Edition 2026.10. Read A2A 101, the first manual [A2A 101](https://aiengineeringfromscratch.com/manual-a2a-101.html) explains the Agent2Agent protocol at v1.0.1, from its purpose to each request and response. It has 28 sections in 7 parts, 7 reference sections, and 31 figures. The PDF is attached below as aiefs-manual-a2a-101.pdf. …View on GitHub ↗