Zero → ML → Deep Learning → RL → LLMs

Learn AI in the right order.

A rigorous, project-driven curriculum built around free Harvard, Stanford, MIT, 3Blue1Brown and PyTorch resources. Your progress is stored locally in this browser.

Start the roadmap →

Curriculum completion

0courses complete

0hestimated studied

0hestimated remaining

Recommended next

CS50P — Python

Foundation · 35 hours

Open course ↗

The curriculum

Step-by-step learning path

Core courses unlock the next stage. Optional resources deepen intuition without blocking your progress.

Build, don’t only watch

Proof-of-learning projects

Finish one small project after each major phase. This is where passive knowledge becomes usable skill.

Learning system

How to use this roadmap

Do not wait until you “finish all mathematics” before touching AI. Alternate theory, implementation and projects.

50% understanding, 50% building

For each mathematical concept, implement a tiny version in NumPy or PyTorch. For each major course, build one project without following a tutorial line-by-line.

concept → derive → implement → experiment → explain

Use AI as a tutor, not autopilot

Ask for explanations, hints and code review. For assignments that are meant to teach implementation, attempt the work yourself before asking for a solution.

struggle a little → ask why → retry → verify

RL comes after ML foundations

Stanford CS234 expects Python, calculus, linear algebra, probability/statistics and foundations of machine learning. Treat those as prerequisites, not bureaucracy.

Keep a learning repository

Create one Git repo with folders for notes, NumPy implementations, PyTorch experiments, solved exercises and projects. Your portfolio grows automatically as you learn.