A 3D version of the classic Rush Hour puzzle game. The decision version of the original 2D game (is there a solution), as well as its generalization to higher dimensions and arbitrary graph topologies, is known to be PSPACE-complete when viewed asymptotically. In laymen's terms, this means the difficulty ceiling is quite high! Try out some algorithmically generated puzzles, as well as some which I crafted by hand.
Open project →
The board game Kalah (part of the Mancala family) was weakly solved in 2000: player 1 wins by 8 with perfect play. But how close can you get to a human-playable strategy that forces a win? Turns out, pretty close! My player 1 bot looks just 2 moves ahead, but still wins ~99.7% of lines. Try playing it for yourself. I also built a 3D game tree visualization with nodes scored by a depth-12 minimax for analyzing your games.
Open project →
A from-scratch blackjack engine, research playground, and card counting trainer. The simulation accommodates different rulesets, card counting strategies, realistic casino shuffles, bankroll management, heat detection, dummy players, and a composition-exact perfect-play solver. You can use it to measure how much edge exists, how to take advantage of it, and what the house does to stop it. Every result is Monte-Carlo with confidence intervals.
Open project →