Equim-chan/Mortal

Built on a deep reinforcement learning model trained end-to-end for Japanese mahjong strategy, with full game rule implementation in Rust for performance-critical evaluation. Provides programmatic APIs alongside comprehensive documentation, enabling both standalone play and integration into mahjong platforms and analysis tools. Training leverages self-play reinforcement learning to develop nuanced decision-making across hand evaluation, discard strategy, and win/loss probability assessment.

1,381 stars. No commits in the last 6 months.