Intelligent healthcare demands objective and scalable methods to evaluate proficiency in robot-assisted surgery (RAS). Existing frameworks rely on subjective observation, leading to bias and inconsistent feedback. This work presents a distributed AI framework for predicting surgical skill and performance using multimodal physiological signals. EEG and eye-gaze data from RAS simulations are analyzed to extract spatiotemporal features reflecting cognitive and motor control. Advanced signal decomposition and visual behavior modeling form a high-dimensional feature space, optimized through distributed feature engineering. Classification and regression models estimate skill levels and performance scores with high accuracy. Real-time inference and monitoring are enabled via Redis, Grafana, and a Streamlit dashboard. Results show precise expertise classification (novice, intermediate, expert) and reliable performance prediction. The framework ensures scalability, interpretability, and seamless integration with surgical simulators, enabling data-driven, real-time assessment of RAS training.