Work / back to index

Poker AI Agent

Game-theory poker that learns not to be exploited

Role

Solo developer

Year

2026

Category

AI / ML

A poker application featuring AI opponents driven by Counterfactual Regret Minimization (CFR). The agent traverses the game tree to compute regret for each action, fold, call, raise, and refines its strategy over time to minimise exploitability in an imperfect-information game.

A real-time opponent-modelling system adapts to player aggression and bluffing frequency, and an event-driven socket layer keeps human and AI game state in sync.

Highlights

  • CFR strategy that approximates Nash equilibrium
  • Adaptive opponent modelling for aggression & bluffs
  • Bluff detection and execution
  • Real-time multiplayer state synchronisation
TypeScriptNode.jsCFRSocket.ioMongoDBReact