In this project, I developed a cutting-edge recommendation system powered by a Field-aware Factorization Machine (FFM) for collaborative filtering. The system leveraged various input signals, including explicit ones like likes and comments, as well as implicit ones like video playtime and skips, to generate personalized recommendations for users on a social media platform.

To ensure efficient and seamless updates, I implemented an incremental training approach, allowing the model to be tuned and updated without the need for complete retraining, thereby minimizing computational resources and downtime.

The trained model was deployed as a highly scalable service, capable of handling substantial throughputs exceeding 50,000 requests per second, ensuring a smooth and responsive user experience.

Collaborating closely with product teams, I conducted extensive A/B testing to optimize user engagement and retention metrics, iteratively refining the recommendation system's performance based on real-world user data and feedback.