Built a Mancala game-playing AI from scratch in Python — game engine, random baseline, and search agents benchmarked over 100-game tournaments.
Implemented minimax and alpha-beta pruning — a ~3.9× speedup at identical win rates that made depth-10 search feasible (plain minimax projected ~2 hrs/game).
Designed a custom heuristic evaluation function crediting end-of-game stone sweeps, raising win rate to 99% at depth 5 (up from 96%).