I’m integrating a new module into my roulette-themed gaming app and need an AI that reliably predicts the next spin’s most probable numbers. The model must learn from three separate data streams—historical wheel outcomes, player betting behaviour logs, and detailed wheel-mechanics data—so it can pick up hidden biases and temporal patterns rather than relying on simple probability tables. Your task is to deliver a trained model plus everything I need to keep it learning as new data arrives. I’m comfortable with Python, TensorFlow/PyTorch, or a lightweight C++/C# inference library; whichever you choose, just expose a clean API that my front-end can call before each spin and return ranked predictions in well under 100 ms. Key deliverables: • End-to-end training pipeline (data cleaning, feature engineering, model training, validation) • Inference module or REST/gRPC service ready to drop into my application • Documentation covering setup, retraining, and integration points • Brief report on achieved accuracy vs. baseline random odds I’ll provide sample datasets and can expand them once we have a working prototype. Looking forward to seeing how you push the edge of game-focused predictive AI.