Turn your regular JavaScript code into a neural net

Handwired converts ordinary JavaScript code into a fully differentiable neural network with no training required, running in a browser without a GPU, according to the project's demos. The Mini OS demo renders a small desktop to regular DOM through 130 layers and 9,909 weights with zero training, while a separate demo searches 100,000 JSON documents by compiling regex matching, JSON parsing and highlighting into WebGPU kernels. The project also lets users unfreeze a compiled program's weights and train them in real time.

Turn your regular JavaScript code into a neural net. No training, exact yet also fully differentiable, runnable in a browser without GPU. Demos Mini OS A small desktop written in JS, transformed with Handwired. Giant lists, animation, stacking, scrolling, drag, input, pictures, all rendering to regular DOM. - weights - 9,909 - layers - 130 - training - 0 Learned gallery Unfreeze a compiled program’s weights & train them in real time . Search 100,000 JSON documents The regex matching, JSON parsing & highlighting are all written in JS, then turned into WebGPU kernels for massive parallelism.