Learn to build, not to watch.

Long-form lessons where the code runs as you read — composed into a plan that fits exactly who you are.

Your first lessons are free — no account needed.

Gradient descent

Walking downhill, one step at a time.

Gradient descent nudges each weight against the slope of the loss. Change the learning rate below and run it — watch the loss fall, or overshoot.

python · runs in your browserlr

w, lr = 2.0, 0.3
for step in range(8):
    grad = d_loss(w)
    w   -= lr * grad
    print(f"loss = {loss(w):.4f}")

Pick a learning rate, press Run.

The difference

Anyone can press play.
Almost no one presses run.

Watching a video feels like progress. Finishing one rarely happens. Here, the only way to move is to do the work — and the work is right there in the page.

Why it sticks

The lesson is the product.

Live Python

It runs in the page.

No notebook server, no local setup. Edit the code, press run, see the output — inside the lesson, in the browser, on any machine.

>>> model.fit(X, y)
epoch 1 · acc 0.71
epoch 8 · acc 0.96 ✓

Videos to sit through.

Prose you read at your own pace — never a play button, never a 12× temptation.

Honest grading

Quizzes graded on the server.

Instant feedback in the page, real scoring behind it. No gaming your own progress.

A plan, not a playlist

Four questions compose your syllabus.

Who you are, where you're stuck, where you're headed, at what pace. Your answers pick a real path — sequenced to your week.

Focus by design

One column. Your progress follows you.

A 68-character reading measure, no clutter, no autoplay — and every completed lesson synced across your devices.

Playgrounds · free, no account

Play with the ideas before you study them.

Interactive demos that run entirely in your browser — no account, no setup. Start with the CNN X-Ray: draw a digit and watch a live convnet light up feature map by feature map, then open any one to see the exact convolution behind it.

Try the CNN X-Ray →All playgrounds

The reader

This is the whole experience.

Prose, live code, inline math and honest quizzes — one surface, no context switching.

Backpropagation, from scratch

The chain rule is the whole trick. If a loss depends on w through an intermediate z, the gradient factors:

∂L/∂w = ∂L/∂z · ∂z/∂w

Read it right to left: the local slope, then the upstream signal. Everything else is bookkeeping — which the code below does for you.

Inline math

KaTeX-rendered, not a screenshot — crisp at any zoom, selectable, searchable.

Honest quizzes

✓The chain rule — correct.

Reading progress

Videos to sit through

100%

Of the code runs in-browser

Questions to your plan

∞

Re-runs, zero setup

The method

Intake, syllabus, practice.

A four-question intake.

Who you are, where you're stuck, where you're headed, at what pace. Nothing graded.

A syllabus from real courses.

Your answers pick a path; the plan is that path's actual courses, sequenced to your week.

A reader built for practice.

Long-form lessons, Python that runs in the page, quizzes graded honestly on the server.

Compose a plan for the engineer you mean to be.

Four questions. A plan from real courses. Free to read, no card — your progress follows you.

Read the first lesson Begin the intake