AI Player for Havannah Board Game

Implemented an intelligent agent for the abstract connection game Havannah, focusing on scalable stochastic search.

MCTS Core: UCB1 selection balancing exploration/exploitation; playout depth constraints for time budgets.
State Representation: Efficient bitset/adjacency forms to accelerate win‑condition detection (fork, bridge, ring).
Heuristics: Progressive bias and move ordering via pattern priors; optional RAVE blending phased out after convergence.
Optimization: Transposition table + rollout policy tuning reduced variance and improved convergence stability.
Benchmarking: >80% win rate against tuned RAVE baseline across randomized openings.

Results: Showcased emergent long‑term strategic structures absent in naive playout agents.

Tech: Python.

Batch simulation mode aggregates playout statistics to improve CPU utilization on multi-core runs.