Extended Master Curriculum · Interactive Edition

The Ultimate "Zero to Hero" AI Roadmap

Math Foundations · Code-First JavaScript ML · Visual Architecture · Python & PyTorch Mechanics
Check off topics as you complete them — progress is saved in your browser.

Progress Tracker 0%
0 of 17 completed Mark topics below as you finish them

Learning Methodology

To build deep, lasting intuition without treating AI as a black box, follow this four-stage progression:

  1. Geometric Intuition & Math — Understand matrices, statistics, and calculus as physical space transformations.
  2. Code-First Intuition (JavaScript) — Build basic neural networks and ML algorithms from scratch visually in the browser without frameworks.
  3. Visual Architecture — Study color-coded diagrams of Deep Learning, Self-Attention, and Transformers.
  4. Line-by-Line Mechanics (Python) — Build autograd engines and complete GPT models from scratch in raw Python and PyTorch.
Phase 1 · Visual & Geometric Math Foundations Zero Code

3Blue1Brown: Essence of Linear Algebra

By Grant Sanderson · YouTube Series (16 videos)

Shows how matrices transform, rotate, and stretch 2D/3D space. Teaches vectors, dot products, determinants, and eigenvectors visually. The single best free linear-algebra intuition resource.

Official playlist → Start: Vectors (Ch.1) →

3Blue1Brown: Essence of Calculus

By Grant Sanderson · YouTube Series (12 videos)

Explains derivatives, rates of change, and the chain rule geometrically. Crucial for understanding how backpropagation updates weights.

Official playlist → Start: The essence of calculus →

Dr. Jon Krohn: Machine Learning Foundations

Comprehensive Math for AI — Hub & 3-Part Curriculum

A complete series bridging visual math to machine learning execution in Python. Start at the main hub, then progress through the three core math pillars.

Main Course Hub → 1. Linear Algebra → 2. Calculus Playlist → 3. Probability & Stats →

StatQuest with Josh Starmer

YouTube Channel · Statistics & ML foundations

Step-by-step visual breakdowns of probability, cross-entropy, Maximum Likelihood Estimation, and foundational ML algorithms.

Channel → StatQuest ML Playlist →

Mathematics for Machine Learning (MML)

By Marc Peter Deisenroth, A. Aldo Faisal, and Cheng Soon Ong

The gold-standard open-access textbook bridging math theory with machine learning concepts.

Download Free PDF → Companion site →

The Matrix Calculus You Need for Deep Learning

By Terence Parr and Jeremy Howard (fast.ai)

A highly readable paper explaining multivariate vector derivatives (Jacobians and Hessians) required for understanding backpropagation.

Read on explained.ai →
Phase 2 · Code-First Intuition (Vanilla JavaScript) No Libraries

Machine Learning Course (From Scratch)

By Prof. Radu Mariescu-Istodor · Visual Algorithms & Fundamentals

Build K-Nearest Neighbors, Decision Trees, and perceptrons from the ground up, starting with a JavaScript drawing app — all without libraries.

Watch Playlist →

Self-Driving Car with JavaScript (Phase 1)

Applied Neural Networks · No libraries

Build a self-driving car system from the ground up. Implement driving mechanics, sensors, and train a neural network visually.

Phase 1 / Full Course Playlist →

Self-Driving Car with JavaScript (Phase 2 – Virtual World)

Genetic Algorithms & 3D Web Environment

Expand the car into a virtual world, applying neuroevolution and complex environmental variables.

Phase 2 Playlist →

Understanding AI (Phase 3 – 15-Part Course)

Deepen the fundamentals of Neural Networks and dimensions

Dive into neural networks, extra dimensions, procedural sounds, and camera inputs using raw JavaScript logic.

Phase 3 Playlist →
Phase 3 · Visual Neural Networks & Transformers Visual Mechanics

3Blue1Brown: Neural Networks (Deep Learning series)

YouTube Series · Chapters 1–7

Animated visualization of multi-layer perceptrons, cost functions, gradient descent, backpropagation, and Transformers.

Full playlist → Start: But what is a neural network? →

TensorFlow Playground

Browser sandbox · No code required

Train neural networks directly in your browser. Adjust hidden layers, learning rates, and activation functions while observing decision boundaries update live.

Open Playground →

Jay Alammar: The Illustrated Transformer

Visual blog post (gold-standard)

The classic color-coded visual guide explaining Embeddings, Self-Attention, Query/Key/Value vectors, and positional encoding.

Read The Illustrated Transformer → Also: Illustrated GPT-2 →
Phase 4 · Code Mechanics: Karpathy's "Zero to Hero" Python & PyTorch

Andrej Karpathy: Neural Networks — Zero to Hero

Former Director of AI at Tesla & OpenAI founding team · Full free course

The definitive line-by-line series. Build autograd engines, language models, and GPT from scratch.

Official playlist → GitHub notebooks →

Lecture 1: Micrograd (Line-by-Line Autograd)

KEY FOUNDATIONAL MILESTONE · Pure Python

Dissecting Micrograd (~150 lines of pure Python) bridges math directly into production software engines.

Watch Lecture 1 → Micrograd GitHub →
The 5 core mechanics dissected line-by-line:
  1. The Value class — wraps numbers to track scalar output (data) and derivative (grad).
  2. Operator Overloading — overrides __add__, __mul__, __pow__ to build a computational graph.
  3. Local Derivatives — explicit calculus derivatives for elementary ops.
  4. Topological Sort & Backprop — reverse-order Chain Rule via .backward().
  5. Building Neurons, Layers & MLP — weighted-sum neurons and multi-layer perceptrons.

Lectures 2–5: Makemore Series (Language Modeling)

Build character-level language models

Progress through Bigrams → Multilayer Perceptrons (MLP) → Batch Normalization → WaveNet-style models.

Start Makemore Series (Lecture 2) →

Lecture 6: Let’s build GPT & NanoGPT

Transformer Architecture & Optimization

Construct a complete generative Transformer matching the original GPT-2 paper from scratch in PyTorch.

Watch Let’s build GPT → NanoGPT GitHub → Let’s build the GPT Tokenizer →

Recommended Study Order (with rough time estimates)

  1. 3Blue1Brown Math Essence Series — ~10–14 h
  2. Jon Krohn ML Foundations & StatQuest — ~15–20 h
  3. Math Papers & Deisenroth Textbook — Reference as needed
  4. Prof. Radu JavaScript AI/ML Series — ~15–20 h (code along visually)
  5. 3Blue1Brown Neural Networks — ~2 h
  6. TensorFlow Playground & Jay Alammar — ~3–4 h
  7. Karpathy Lecture 1 (Micrograd) — ~8–12 h (watch + re-implement from scratch)
  8. Karpathy Makemore series (Lec 2–5) — ~10–15 h
  9. Karpathy Let’s build GPT + Tokenizer — ~15–25 h

Propose a Resource / Suggest Sequence Change

Found a better resource or a smarter order? Share it below.

Thanks — your proposal was recorded. I’ll review it soon.