Decode
Private beta

Ask a question.
Get a video that teaches.

Type a question like “How does backpropagation work?” and Decode makes you an animated video that walks through it properly, with narration and visuals built to teach. Not a wall of text to skim.

A generated lesson · self-attention → transformer → gradient descent1920 × 1080
Narration is the clock. Every visual lands on the words that name it.

One question, and Decode does the rest.

01Ask

Type your question. Or drop in a paper, a doc, the chapter you’re stuck on. One message is all Decode needs.

02Watch it build

Decode plans the lesson, writes the narration, then animates each scene. It tells you what it’s doing while it works, so you’re never staring at a spinner.

03Watch it teach

A few minutes later you have a film that actually explains the thing, start to finish, timed to its own narration.

A film that teaches. Not a chatbot that talks.

No stock footage with a voice on top. Every scene draws the idea itself — the parts, how they connect, the step where it clicks — and lands on the words that name it. These four frames came out of one generated lesson.

The transformer stack drawn as blocks: embedding, attention, add and norm, feed forward, with residual connections arcing past each
The transformer stack, with the residual connections that skip around each block.
Five token chips with arcs between them; a thicker arc from “it” back to “cat”
Attention across a sentence — a thicker arc is a stronger weight.
A matrix multiply Q times K-transpose with one row of Q, one column of K and their output cell highlighted together
Scores = Q × Kᵀ: one row dotted with one column makes one cell.
Contour rings of a loss landscape with a path of steps walking to the minimum at the centre
Gradient descent walking downhill to the minimum, one step per spoken beat.

Bring a question. Leave with a film.

Decode — ask a question, get a video explanation