rohitg00/ai-engineering-from-scratch
A free hands-on curriculum that teaches AI by having you build every piece from scratch, from math to shippable agents
This repository is a free, open-source curriculum that teaches AI engineering by having learners implement everything by hand, from math foundations to autonomous agents. It spans 20 phases and 511 lessons, and every lesson ends with a reusable artifact like a prompt, skill, agent, or MCP server. Installing the learning skills into a coding agent turns your terminal into an interactive tutor that builds a personalized study plan and teaches one lesson at a time.
What it does
- The 511 lessons across 20 phases cover setup, math foundations, classical ML, deep learning, transformers, large language models, multimodal AI, agents, and production infrastructure, written in Python, TypeScript, Rust, and Julia.
- Every lesson follows the same six-step structure: motto, problem, concept, build-it-from-raw-math, use-it-in-a-production-library, and ship-it as a reusable artifact.
- Running `npx skills add` installs learning skills into agents like Claude Code or Codex, which then run a ten-question placement quiz, save a personalized plan to LEARNING.md, and teach lessons interactively one at a time.
- Separate focused paths exist for learners who only want Model Context Protocol (17 lessons) or Agent Skills (5 lessons), each tracked in its own progress file.
- The 20-phase core curriculum is also compiled by CI into a six-volume EPUB/PDF book series attached to every GitHub release.
Why it matters
It pushes learners past merely calling AI APIs toward implementing the actual mechanics, like attention and the agent loop, from raw code, so they can debug and reason about systems they build rather than treat them as black boxes. Because each lesson produces a working artifact, studying the curriculum simultaneously builds a practical portfolio of real tools.
Terms in this repo
- MCP (Model Context Protocol) · a standardized protocol that lets AI models access external tools and data
- Agent Skills · standardized SKILL.md instruction files that an AI agent reads to perform a specific task
- LoRA/QLoRA · lightweight fine-tuning techniques that train a small set of added parameters instead of the whole model
- RAG (retrieval-augmented generation) · a technique where the model retrieves external documents before answering, using them as grounding
- agent loop · the repeating cycle where an AI reads tool results and decides its next action
Repository description (English)
Learn it. Build it. Ship it for others.
Open on GitHubTrending repos
- vorssaint/vorssaint-utilsOne free menu bar app replaces a dozen paid Mac utilities
- MadsLorentzen/ai-job-searchAn open-source framework that runs your entire job search through Claude Code, built by someone who actually got hired using it
- AgriciDaniel/claude-obsidianAn open-source second brain that lets Claude Code file your notes into Obsidian while you keep full ownership of the files
- openclaw/openclawA personal AI assistant that lives on your own devices and talks to you through the messaging apps you already use
- tashfeenahmed/freellmapiAn open-source router that stitches 34 free LLM tiers into one API key so your app never hits a rate limit
- Alishahryar1/free-claude-codeA local proxy that lets coding AI agents run on 49 free or cheap model providers instead of one paid service
- freestylefly/awesome-gpt-image-2A library of 532 reverse-engineered prompts that turn GPT-Image2 into a predictable image-making tool
- block/buzzAn open-source workspace where humans and AI agents chat, code, and review in the same rooms
Latest from METAL LAB
- ChatGPT Work now turns spoken thoughts into finished documents
- 100-Rally Match Against World No. 15... The Robot on Court Was Unveiled Three Days Earlier
- Robot 100m in 9.32 Seconds Beats Bolt in Heats, But Foam Pads Wait at the Finish Line
- NVIDIA in talks to invest in Perplexity at $30 billion-plus valuation
- Inside Sakana AI — the $2.7 billion company a Transformer author built in Tokyo