IRumAI: Reinforcement Learning for Indian Rummy

IRumAI is a novel reinforcement learning agent designed for Indian Rummy, utilizing Proximal Policy Optimization (PPO) and a dual-branch convolutional architecture. It achieves a 53.9% win rate against the strongest search-based opponent, demonstrating significant performance with a rapid inference time of 0.33 ms per action, which is over 7,000 times faster than existing heuristic methods. This advancement is crucial for practitioners as it offers a new approach to handling complex hidden-information games efficiently without relying on explicit search techniques.