Python in your browser
Pyodide runs NumPy, matplotlib, and your code right in the page. No setup, no Colab tab dance.
v0.4 · curriculum 15 phases · 86 lessons · in-browser code labs
Learn how LLMs
actually work.
A ground-up curriculum — math, code, intuition. Run Python in the browser, visualize attention, and ship working transformers. Free, open, self-paced.
Built for technical learners
Everything you need to ship a transformer.
AI tutor (BYOK)
Anthropic, OpenAI, Google, OpenRouter, Ollama, DeepSeek — bring your own key. We never spend on AI, you do.
Interactive visualizations
Drag arrows, slide softmax temperature, step through matmul — feel the math before you see it.
XP, streaks & leaderboards
Daily streaks. Phase mastery badges. A global leaderboard if you want it. All optional, never patronizing.
The full curriculum
14 phases. Violet to magenta.
Each phase has its own color, its own arc. Start at the foundations or jump in where you're stuck.
P00
Foundations: Math & Python from Zero
No math background? Start here. From arithmetic to gradients and Python from the first line — everything Phase 1 quietly assumes.
P01
Math & ML Refresher
The vector + gradient + probability foundation every LLM idea rides on.
P02
Deep Learning Foundations & PyTorch
Write training loops you understand down to autograd. Debug exploding gradients with confidence.
P03
NLP Fundamentals
From word vectors to attention — and why attention won.
P04
Transformers Deep Dive
Implement multi-head attention and a full decoder-only Transformer from a blank file.
P05
Modern LLM Architectures & MoE
Llama 4, Qwen3, DeepSeek V3/R1 — what's actually inside the frontier.
P06
Pre-Training — Scaling, Data, Distributed Systems
Estimate compute-optimal trade-offs; use FSDP and parallelism that doesn't OOM at 3am.
P07
Fine-Tuning & Alignment
Take a base model to instruction-following with the right algorithm for your budget.
P08
Prompt Engineering & Reasoning
From ad-hoc prompting to a real vocabulary — CoT, self-consistency, budget forcing, reasoning models.
P09
Retrieval-Augmented Generation
Build a production-grade RAG that beats naive cosine-similarity on your own eval.
P10
Agents, Tool Use, and MCP
Multi-tool, stateful, durable agents. Expose your tools to the world via MCP.
P11
Evaluation
Trustworthy evals before you trust any output. Goodhart's law in practice.
P12
Deployment & Inference Optimization
Serve at >1000 tok/s/GPU. Quantize down to consumer hardware without losing your benchmarks.
P13
Safety, Alignment & Interpretability
Red-team your own model. Reproduce a basic SAE. Understand current limits of mech interp.
P14
Research Track — Reading, Reproducing, Publishing
Read 1 paper/week with full understanding. Reproduce one paper per quarter. Ship a preprint.