Don't read about machine learning.
Watch it happen.
Every concept here — from a Python list to multi-head attention — comes with something you can drag, step through and break. Move the line and watch the error change. Turn the learning rate up and watch it diverge. The maths and the code are here too, but the picture comes first.
Six sections, nineteen modules, in the order they build on each other.
Nothing here assumes knowledge it hasn't taught you. Start at the top if you're new, or jump to whichever module is the gap in what you already know.
Prefer a map? Explore the classical ML knowledge graph — every algorithm as a node you can open and play with.
Nine layers, same order, every time.
Intuition before formalism — always. You never meet an equation before you know what problem it solves, and you never meet code before you've seen the mechanism it implements.
The problem
What breaks or becomes impossible without this idea. You feel the pain before you see the cure.
The intuition
The whole concept in three sentences, plus one analogy from software you already understand.
The mechanism
The interactive visualization. Real, small numbers you can trace by hand — and change yourself.
The math
Now the equations, with every symbol defined the moment it appears and a plain-English line under each.
From scratch
Forty lines of NumPy or less, using the exact numbers from the visualization. Runnable as-is.
In production
The real PyTorch, scikit-learn or Pandas call, and what the library does for you that your version didn't.
When to use it
And when not to. This is the layer that separates knowing how from knowing when.
Pitfalls
The mistakes everyone makes, written as a runbook: here's the symptom, here's the fix.
Check yourself
One question you should be able to answer now. It isn't answered for you — that's the point.
Optimisation is the best first lesson.
Gradient descent is the algorithm underneath almost every model in this track, and it's the clearest possible demonstration of what this site is for: turn the learning rate up and watch a working model fly apart in front of you.