Handwired: compile plain JavaScript into exact, differentiable neural nets
TypeScript toolchain that turns if-else, loops and state into matmul + ReLU, producing neural nets that exactly match the original JS and can run on WebGPU.
Why it matters
A widely-seen bridge between deterministic code and deep learning: ordinary programs compile into exact neural-net twins that stay fully differentiable, so compiled code can be trained, fine-tuned, or fused with other models after compilation.
What you could build with it
An indie developer or dev-tools startup could build a differentiable rules engine for e-commerce: merchants paste pricing and discount logic as plain JavaScript, get back exact neural twins, then fine-tune only the edge cases on real order data instead of rewriting the logic.
Does it hold up?
Early but real: the author's live demos plus an independent dev.to hands-on test confirm compiled nets exactly match the original JS; still an experiment to play with, not a production toolchain.
Built with Handwired
- Turned into a neural net, my code learned every price and none of the rulesdev.to · Independent hands-on test: fed a five-rule shipping function and its neural twin 16M inputs with zero disagreement, then showed what breaks when the twin is trained on new price lists.
Learn more
First spotted on github: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71NVIDIA Nemotron-pretraining-Indic-v1: 2.8T-token open Indic pretraining corpusNVIDIA with IIT Madras's AI4Bharat released a ~2.8 trillion-token pretraining corpus covering 22 Indian languages plus…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.