Hi Product Hunt! 👋 A few years ago, I open-sourced a sequence-prediction training project on GitHub (predict_Lottery_ticket). To my surprise, it crossed 1k stars — turns out a lot of people are curious about applying neural networks to lottery data, even knowing it can't actually be predicted. But there was a problem: most people who wanted to try it didn't know how to train models, and nobody wants to rent a GPU server just to experiment with this. So I rebuilt the whole thing as a browser-based tool. The entire training process runs locally via TensorFlow.js — no server, no GPU rental, no setup. Just open the page and hit "Train". What it does: • Trains LSTM / Transformer models fully in your browser • Visualizes the full probability distribution for every number (with heatmaps), not just a single "predicted" number • Supports 5 lotteries across CN / HK: SSQ, DLT, QLC, QXC, HK Mark Six • Community feature — users can share their picks and see what others chose • Stack: Next.js + TensorFlow.js + Supabase An honest note: this is NOT a prediction tool, and never will be. Lottery draws are independent random events. The value here is learning how sequence models behave on noisy, high-entropy data — and seeing firsthand why they can't predict randomness. Which is itself a pretty interesting lesson. Would love to hear from anyone who's played with sequence models on noisy datasets. Happy to answer implementation questions. Thanks for stopping by! 🎲

AI Spinach — Train ML in Your Browser

Train neural nets in your browser — no setup, no server

Train neural nets in your browser — no setup, no server