⬢github TypeScript · 1.9K ★ +993 since we first saw it · pushed 4 d ago · MIT
chenglou/handwired
Turn your regular JavaScript code into a neural net.
Handwired transforms ordinary JavaScript programs—conditionals, loops, state—into neural-network form: just matrix multiplications and ReLU, exactly preserving results with no training needed. Since the output is only matmuls, it stays fully differentiable, runs on JS or WebGPU, and enables experiments like training the 'weights' of hand-written code or running traditionally single-threaded JS with GPU-level parallelism.
Why now: It's a viral new experiment (1.7k stars within two days) from a well-known developer, with live demos showing surprising tricks like millisecond JSON search over 100k documents via compiled JS-as-matmuls.
Who it is for: JavaScript developers and ML-curious engineers who want to play with the boundary between code and neural networks.
Stars over our 264 snapshots: 865 to 1.9K, since 5 d ago.
Where people talked about it
- ⬢github new repos, most starred 5 min ago
- ⬢github new repos, most starred 6 h ago
- ⬢github new repos, most starred 1 d ago
- ⬢github new repos, most starred 2 d ago
- ⬢github new repos, most starred 3 d ago
- ⬢github new repos, most starred 4 d ago
- ⬢github new repos, most starred 5 d ago
API: https://socialmediatrends-api.osmike.com/v1/repos/chenglou/handwired