Real-Time Content Recommendation Engine with Machine Learning

Design a sophisticated recommendation system using Python's scikit-learn and pandas that predicts user content preferences for a streaming platform. Implement collaborative filtering with hybrid recommendation techniques, incorporating both user behavioral data and content metadata. The system should handle dynamic updates, support real-time learning, and provide recommendation confidence scores. Include performance optimization strategies for handling millions of user interactions with less than 50ms latency.