Hokm Information

I built an Iranian card game, Hokm, and one of the most interesting technical challenges has been the AI. At first glance, Hokm looks like a rules problem. In practice, it is a partnership and inference problem. Strong players are not only following rules. They develop style, tempo, signaling habits, and risk preferences that their partner gradually learns to read. That makes the game much harder to model than a simple trick-taking engine. The current bot is a rule-based AI system, not a learned model. It builds a structured view of the hand, derives public and inferred information, generates candidate actions across multiple decision layers, and selects a move through deterministic priorities and heuristics. This works well for transparency, control, and fast iteration, but it still has an important limitation: it can play logically without truly adapting to human style. The next step is to add a machine learning layer on top of this architecture. I am building an analytics pipeline around gameplay events and decision features, with BigQuery as part of the data path, so I can evaluate where heuristic decisions align with strong human play and where they do not. The goal is not just to make the bot stronger, but to make it more expressive: more sensitive to context, more responsive to partnership patterns, and eventually more capable of reflecting different styles of play. To support that, I have been asking experienced Hokm players to complete at least one full set of rounds in the app. That gameplay data helps create the behavioral footprint needed to move from hand-crafted logic toward data-informed decision making.We provide you with the shoppers based on geographic locations, saving your time, packaging costs and more sustainable sales. Our Machine Learning helps you to save time to answer repeatable questions from different sources. Therefor e you don't need to answer all questions coming from shoppers.