Work / back to index

Risk-Aware Poker AI

Max-EV play, with a risk manager on top

Role

Solo developer

Year

2025

Category

AI / ML

Risk-Aware Poker AI addresses the risk-management gap in prior poker-AI work by adding an explicit risk layer to the decision engine. An analyst picks the max-EV action; a 'boss' risk layer then applies a hard guardrail (force-fold when the bet-to-bankroll ratio crosses a threshold), a soft risk utility that ranks actions by EV minus a penalty on risk and variance, and an optional bankroll floor.

A Flask API exposes the decision endpoint to a React frontend.

Highlights

  • Hard guardrail force-folds on bankroll risk
  • Risk-adjusted utility: EV − λ·(risk + variance)
  • Configurable bankroll floor
  • Sweepable thresholds for experimentation
PythonFlaskReactNumPy