DEV Community

TengLongAI2026
TengLongAI2026

Posted on

473 Lessons, Zero Fluff: The Open-Source AI Engineering Degree Taking Over GitHub

#opensource

473 Lessons, Zero Fluff: The Open-Source AI Engineering Degree Taking Over GitHub

TL;DR: 25,153★ open-source curriculum (473 lessons, 20 phases, ~320 hours) covering everything from linear algebra to autonomous agent swarms. Every lesson produces a reusable artifact. MIT license.

Summary

There's no shortage of "learn AI" resources. YouTube tutorials, paid courses, bootcamps. Most teach you to call APIs. Few teach you what happens inside.

ai-engineering-from-scratch (25K★) is different. It's a complete, structured curriculum with 473 lessons across 20 phases. You build everything from first principles — backprop, tokenizer, attention mechanism, agent loop — before touching any framework.

The Scale

Metric Number
★ Stars 25,153
Lessons 473
Phases 20
Languages Python, TypeScript, Rust, Julia
Reusable artifacts 473 (1 per lesson)

The 20 Phases 

Phase  0 — Setup & Tooling (12 lessons)
Phase  1 — Math Foundations (22)
Phase  2 — ML Fundamentals (18)
Phase  3 — Deep Learning Core (13)
Phase  4 — Computer Vision (28)
Phase  5 — NLP: Foundations to Advanced (29)
Phase  6 — Speech & Audio (17)
Phase  7 — Transformers Deep Dive (14)
Phase  8 — Generative AI (14)
Phase  9 — Reinforcement Learning (12)
Phase 10 — LLMs from Scratch (22)
Phase 11 — LLM Engineering (17)
Phase 12 — Multimodal AI (25)
Phase 13 — Tools & Protocols (23)
Phase 14 — Agent Engineering (42)
Phase 15 — Autonomous Systems (22)
Phase 16 — Multi-Agent & Swarms (25)
Phase 17 — Infrastructure & Production (28)
Phase 18 — Ethics, Safety & Alignment (30)
Phase 19 — Capstone Projects (55)
Enter fullscreen mode Exit fullscreen mode

What Makes It Special

1. Build It / Use It Split

Every lesson follows a six-beat structure. First you implement the algorithm from scratch. Then you run the production version. You understand PyTorch because you built the smaller version first.

2. Every Lesson Ships Something

Other curricula end with "congratulations, you learned X." Each lesson here ends with a reusable tool — prompts, skills, agent loops, MCP servers.

Install all 473 with one command: python3 scripts/install_skills.py

3. Built-in Agent Skills

The curriculum includes two AI agent skills:

  • /find-your-level — 10-question placement quiz
  • /check-understanding <phase> — Per-phase quiz

Quick Start 

git clone https://github.com/rohitg00/ai-engineering-from-scratch.git
cd ai-engineering-from-scratch
python phases/01-math-foundations/01-linear-algebra-intuition/code/vectors.py
Enter fullscreen mode Exit fullscreen mode

FAQ

Q: Need a GPU? A: Most early phases run on CPU.

Q: Prerequisites? A: Basic programming ability.

Q: Free? A: MIT license. Free forever.

Why This Matters

Most people learn AI in pieces. A paper here, a tweet there. The pieces rarely line up.

This curriculum provides the spine. By the end, you've built an agent loop from raw Python, trained a transformer from scratch, and deployed a multimodal system to production.

More importantly: you own the artifacts. 473 prompts, skills, agents, and MCP servers that you understand because you built them.

Started the curriculum? Drop your phase below. Follow for more open-source deep dives.

Tags: #ai #opensource #learning #python

Top comments (0)