Project Mosaic
Zero-install decentralized AI compute in your web browser
6 followers
Zero-install decentralized AI compute in your web browser
6 followers
Project Mosaic replaces expensive centralized data centers with a browser-native decentralized AI compute grid. Using WebGPU, everyday consumer devices can collaboratively train and fine-tune foundation models on the edge with zero setup—no Python or CUDA drivers required. Key Features: • Zero-install WebGPU & Wasm fallback • 100% Zero-Leakage Edge Privacy • Gamified AI training via Retro Arcade
Hey Product Hunt community! 👋
I’m Kanishk, Founder of Project Mosaic.
Right now, the AI industry is running headfirst into a massive bottleneck: multi-billion-dollar data centers that burn city-scale electricity, cost $100M+ to build, and centralize everyone's private data. Meanwhile, billions of dollars of high-performance consumer silicon (Apple Silicon, RTX GPUs, modern NPUs) sit idle on our desks.
We asked a fundamental question: What if we could train massive foundation models across a decentralized swarm of everyday browsers instead of renting giant server farms?
Today, we are excited to launch Project Mosaic V1:
Zero-Install WebGPU: Train lightweight 4-bit LoRA parameter updates locally inside your browser with automatic Wasm fallback.
100% Zero-Leakage Privacy: Data stays in device memory; only scrambled mathematical parameter summaries leave the machine.
EcoMind Built-In Search: A sustainable search fallback that resolves out-of-domain queries without blowing up memory limits.
Gamified Compute (Arcade Hub): Anyone can contribute compute micro-batches simply by playing retro web games like Neural Super-Highway Racer.
You can test it live in 1 click with zero setup or login (just hit "⚡ Continue as Guest"): 🌐 App:[https://www.projectmosaic.in](ht...
We’d love to hear your raw feedback, stress-test results, and questions. I'll be hanging out in the comments all day to answer your technical questions about our WebGPU architecture! 🚀