M20 / LABORATORY · FLAGSHIP

The Sandbox. Chain anything
to anything.

Drop nodes onto the canvas. Wire them. Image source → convolution → pooling → embedding → output. Every node runs live. The whole graph re-computes on every drag. This is what every CV pipeline looks like underneath.

EXPERIMENT 10.01 · SANDBOX.GRAPH NODES · 7 EDGES · 6 EXEC TIME · 3.2ms

INPUT

PROCESSING

OUTPUT

STATUS · RUNNING

DRAG nodes to rearrange

CLICK port to start wire

CLICK port to finish wire

DBL-CLICK node to delete

This is the whole point → Modern ML systems are graphs of differentiable ops. Frameworks like PyTorch and JAX let you assemble these graphs in code; what you're seeing here is what they assemble underneath. Every output of every node is a tensor; every edge a tensor flow.