28 Card Game — AI-Powered Online Play Live
28cardgame.com is the definitive online home for 28 — the popular South Asian trick-taking card game. The client wanted more than a basic multiplayer lobby: they wanted AI opponents that could challenge experienced players, immersive 3D card visuals that felt like sitting around a real table, and a platform that could host thousands of concurrent games without lag.
I designed and built the full stack. A TensorFlow-trained neural network powers the AI opponents, learning bidding strategy, trump selection, and trick play from hundreds of thousands of recorded games. Three.js renders an interactive 3D card table with fluid animations for dealing, bidding, and trick-taking. A Node.js real-time backend orchestrates multiplayer lobbies, matchmaking, game state synchronisation over WebSockets, and leaderboards.
Highlights
- AI opponents trained with TensorFlow — the model learns bidding aggression, trump timing, and partner-play signals from historical game data, producing opponents that adapt to different skill levels rather than following scripted rules.
- 3D card table rendered in Three.js / WebGL — smooth dealing, fanning, and trick animations with dynamic lighting and camera angles that give the feel of a physical card game in the browser.
- Real-time multiplayer over WebSockets — players can create or join rooms, invite friends, or get matched automatically. Game state stays synchronised across all four seats with sub-second latency.
- Full 28-game rules engine — point-based bidding, hidden trump declaration and reveal, trick evaluation, and automatic score tallying with support for common regional rule variants.
- Leaderboards, player profiles, and match history — Elo-style rating system that tracks wins, points, and streaks across ranked and casual modes.
- Responsive across devices — the 3D table scales from desktop monitors down to mobile screens with touch-friendly card selection and gesture-based interactions.