About this project

AI Engineering from Scratch is a free, MIT-licensed curriculum and reference manual that teaches AI engineering from first principles rather than API wrappers. The repository contains 523 lessons organized into 20 sequential phases, estimated at roughly 342 hours of work, with implementations in Python, TypeScript, Rust, and Julia. The curriculum builds bottom-up: setup and tooling, math foundations, classical ML, deep learning, vision, NLP, speech/audio, transformers, generative AI, reinforcement learning, LLMs from scratch, LLM engineering, multimodal systems, tools and protocols (including Model Context Protocol), agent engineering, autonomous systems, multi-agent swarms, infrastructure/production, ethics/alignment, and capstone projects. Each phase depends on earlier ones, and each lesson follows a consistent structure: a one-line motto, a concrete problem, conceptual diagrams, a "Build It" section implementing the algorithm in raw math without frameworks, a "Use It" section running the same idea through PyTorch or scikit-learn, and a "Ship It" section that leaves a reusable artifact — a prompt, skill, agent, or MCP server. Entry paths are goal-oriented rather than forcing linear reading. There are fast routes for newcomers, Python developers wanting ML foundations, people building production LLM applications, agent builders, MCP developers, Agent Skills authors, and Claude certification candidates. A ten-question placement quiz maps existing knowledge to a starting phase and writes a personalized plan to LEARNING.md. The repo also functions as an AI-tutor package: `npx skills add rohitg00/ai-engineering-from-scratch` installs skills (start-learning, learn, course-guide, learn-mcp, learn-agent-skills, claude-certification, find-your-level, check-understanding) that work with Claude Code, Codex, and other SKILL.md-compatible hosts. Progress is persisted locally in files like LEARNING.md, MCP-LEARNING.md, and AGENT-SKILLS-LEARNING.md. The 20-phase core compiles into a six-volume EPUB/PDF book series, built by CI and attached to GitHub releases. The certification academy is independent study material based on public exam objectives, not affiliated with Anthropic. The README notes reported readership of 114,584 readers and 181,995 page views in the last 30 days. Landing pages are translated into multiple languages, with English as the canonical source.