I propose building a Python-based TensorFlow Lite model for short video recommendations. Leveraging user engagement data, the model will tailor recommendations by excluding watched videos and incorporating popular content. Integration with Firebase Firestore ensures efficient data management, while Firebase ML facilitates seamless deployment. The workflow encompasses data retrieval, similarity computation, popular content integration, and model training. Deliverables comprise the TensorFlow Lite model, Firebase integration, and detailed documentation. Emphasis will be placed on Flutter compatibility, mobile optimization, and robust privacy protocols to ensure a user-centric and secure recommendation system.