AI Engineering from Scratch

Build transformers, LLMs, and AI agents from first principles - verified by graded code, running entirely in your browser.

$20 lifetime access to graded exercises, an AI tutor, and a verified certificate. The curriculum itself is free, based on the open MIT course by Rohit Ghumare.

Why this course is different

  • Build, don't just watch. Every lesson has a graded in-browser coding exercise. Your code runs against real automated tests inside the browser via Pyodide (Python-in-WASM) - no installs, no cloud account.
  • Verified by machine, not vibes. The certificate is earned by passing autograded tests, not clicking through slides. Employers can trust it.
  • From first principles. You implement transformers, attention mechanisms, backpropagation, and LLM inference from scratch - in Python, in your browser.
  • AI tutor included. Bring your own API key (Anthropic, OpenAI, or Gemini) and get a context-aware tutor that knows exactly what lesson you are on and never gives away the solution.
  • No GPU needed. All 20 phases run on browser WASM. Deep-learning phases use numpy-level implementations so any laptop works.
  • $20 once, lifetime access. No subscription, no per-lesson fees.

20-Phase Curriculum (260+ lessons)

Each phase contains multiple lessons. All reading is free. Graded coding exercises unlock with the $20 one-time payment.

  1. Phase 0 - Setup and Tooling: Environment setup, Python fundamentals, toolchain for AI engineering.
  2. Phase 1 - Math Foundations: Linear algebra, calculus, probability, statistics, information theory, and norms - all implemented from scratch.
  3. Phase 2 - ML Fundamentals: Supervised and unsupervised learning, gradient descent, decision trees, SVMs, clustering built from first principles.
  4. Phase 3 - Deep Learning Core: Backpropagation, neural networks, activation functions, batch normalization, dropout - implemented in pure numpy.
  5. Phase 4 - Computer Vision: Convolutions, CNNs, image classification, object detection architectures built from scratch.
  6. Phase 5 - NLP Foundations to Advanced: Tokenization, embeddings, word2vec, sequence models, attention mechanisms.
  7. Phase 6 - Speech and Audio: Audio processing, spectrograms, speech recognition fundamentals.
  8. Phase 7 - Transformers Deep Dive: Multi-head attention, positional encoding, encoder-decoder, the full transformer architecture - built from scratch.
  9. Phase 8 - Generative AI: VAEs, GANs, diffusion models, generative techniques from first principles.
  10. Phase 9 - Reinforcement Learning: MDPs, Q-learning, policy gradients, RLHF fundamentals.
  11. Phase 10 - LLMs from Scratch: Pre-training, tokenization (BPE), causal attention, GPT-style language model implementation.
  12. Phase 11 - LLM Engineering: Fine-tuning, RLHF, inference optimization, quantization, serving LLMs in production.
  13. Phase 12 - Multimodal AI: Vision-language models, cross-modal attention, multimodal embeddings.
  14. Phase 13 - Tools and Protocols: Function calling, tool use, MCP (Model Context Protocol), structured outputs.
  15. Phase 14 - Agent Engineering: ReAct agents, planning, memory, tool-using agents built from scratch.
  16. Phase 15 - Autonomous Systems: Agentic loops, long-horizon planning, autonomous decision-making systems.
  17. Phase 16 - Multi-Agent Systems and Swarms: Multi-agent coordination, swarm intelligence, agent communication protocols.
  18. Phase 17 - Infrastructure and Production: MLOps, model deployment, monitoring, scaling AI systems.
  19. Phase 18 - Ethics, Safety, and Alignment: AI safety fundamentals, alignment techniques, responsible AI engineering.
  20. Phase 19 - Capstone Projects: End-to-end AI engineering projects integrating skills across all phases.

Frequently Asked Questions

What is AI Engineering from Scratch?

A 20-phase, 260-lesson course teaching you to build AI systems - transformers, LLMs, agents, computer vision models, and more - from first principles. All coding runs in your browser via Pyodide (Python-in-WASM). No installs. Based on the open MIT curriculum by Rohit Ghumare.

Is the course content free?

Yes. The full 20-phase reading curriculum is freely accessible to everyone. The $20 one-time payment unlocks graded exercises, the AI tutor, and the verified completion certificate.

What does the $20 lifetime access unlock?

Three things: (1) graded in-browser coding exercises with an autograder that checks your solution against real automated tests, (2) an AI tutor (bring your own API key for Anthropic Claude, OpenAI, or Gemini) that answers questions in context of each lesson without spoiling solutions, and (3) a verified completion certificate earned by passing all graded exercises.

Is the certificate verified?

Yes. You earn it by passing machine-graded coding exercises, not by watching videos. Every graded lesson has tests that your solution must pass. The autograder is the same one that verified the lesson's reference solution. This makes the certificate verifiable and meaningful.

Do I need a GPU?

No. All 20 phases run in the browser via Pyodide (Python compiled to WebAssembly). Numpy, scikit-learn-style libraries, and custom implementations run on any modern laptop - no GPU, no cloud compute, no local Python installation required.

What AI topics does this cover?

Math and statistics, machine learning fundamentals, deep learning, computer vision, NLP, speech, transformers, generative AI, reinforcement learning, LLMs from scratch, LLM engineering, multimodal AI, tool use and MCP, agent engineering, autonomous systems, multi-agent swarms, MLOps, AI safety and alignment, and capstone projects.