← Discover
Discover · GitHub · AI Engineering from Scratch↑ +2,169 today

Artificial Intelligence Engineering Training from Scratch

Introduction to artificial intelligence engineering from scratch: 485 lessons, 20 phases and approximately 320 hours of content. He teaches every algorithm, from linear algebra to autonomous agents, first with mathematics and then by writing code by hand. This course, which covers Python, TypeScript, Rust and Julia languages, is completely free and open source.

Updates

  • September 8, 2026: Stars 52,433 → 52,978, latest release v2026.09 (September 7, 2026).
  • September 5, 2026: Stars 49,881 → 52,433, latest release v2026.08 (August 10, 2026).
  • August 27, 2026: Stars 46,486 → 49,881, latest release v2026.08 (August 10, 2026).
  • August 11, 2026: Stars 45,404 → 46,486, latest release v2026.08 (August 10, 2026).

What do you learn?

  • You set up the basics such as backprop, tokenizer and attention manually, starting with the math.
  • You develop end-to-end, from the agent loop to the MCP server.
  • Each lesson leaves a reusable output: a prompt, an ability, or an agent.
  • When you use PyTorch, you already know what's going on behind the scenes.

How it works

Each lesson follows the same cycle: analyze the problem, do the math, write the code, run the test, and store the output. There are no five-minute videos or copy-paste deployment processes. Everything runs on your own computer.

How to use it with AI?

🤖 Paste this into your AI agent (Claude Code · Codex · Antigravity)

I'll give you the content of a 20-phase, 500+ lesson open source curriculum called 'ai-engineering-from-scratch' that teaches AI engineering from scratch. Review the README and the roadmap and it gives me a personal study plan, in Turkish, divided into weeks: which phase should I start from, what are the prerequisites and in what order should I proceed? My level: [eg. beginner].

Related dictionary terms

TokenizerAI EngineeringAttentionPromptMCPOpen Source
Who it is forThose who want to learn AI fundamentally rather than superficially
DifficultyIt progresses from the beginning forward, phase by phase
Format485 lessons · 20 phases · ~320 hours
languagesPython, TypeScript, Rust, Julia
FeeFree · open source (MIT)
License: MIT · you can freely use, adapt and share content and code.

Links

TreScout did not build this tool · we found it in GitHub trends and wrote it up. This page describes the repository as of 2026-05-26: The star count and our text belong to that day, the repository may have changed since. Check the repository link for the current state. This page was machine-translated from the Turkish original · the Turkish version prevails.