Inside Amaka AI: how an equation becomes an animation

The hard part of visual teaching is not rendering. It is deciding what to draw, in what order, and when to stop talking. Amaka AI treats that as a planning problem before it treats it as a graphics problem.

Four stages

1. Decompose the concept

A question like why does a heavier object not fall faster is broken into the smallest set of ideas that have to land in sequence: force, mass, acceleration, and the cancellation between them. Each becomes a section with its own visual goal.

2. Choose the visual grammar

Every section is mapped to visual primitives — an axis, a vector, a body, a label, a highlight — drawn from a fixed vocabulary. Constraining the vocabulary is what keeps a hundred different explanations looking like one coherent teacher rather than a hundred different clip-art styles.

3. Score the timing

Strokes are timed against the narration so the label arrives as the word is spoken, not three seconds before. This is the difference between a diagram appearing and a diagram being drawn at you.

4. Render and stay interruptible

The board renders progressively and holds its state, so a follow-up question can attach to what is already on screen instead of resetting it.

Why not just generate a video?

Because a video cannot be asked a question. The moment a learner does not follow a frame, a rendered video has nothing to offer them but the scrub bar. Keeping the explanation as live, structured state — rather than pixels — is what makes the follow-up possible.