14 min • 3 lectures
From Sun Team to Greg. This course provides a technical briefing on the architecture of recommendation systems specifically designed for mobile content applications. It addresses the fundamental shift from traditional search-based discovery to automated push-based feeds. On mobile, engineers must account for limited screen real-estate and high session frequency by prioritizing implicit feedback signals such as dwell time, scroll depth, and tap-through rates. The curriculum breaks down the recommendation pipeline into three distinct phases: candidate generation, ranking, and re-ranking. Each stage is analyzed through the lens of mobile latency constraints, ensuring the entire loop completes in under 100 milliseconds to match the speed of user interaction. The technical instruction begins with candidate generation, exploring the trade-offs between collaborative and content-based filtering. It covers two-tower model architectures and the use of Approximate Nearest Neighbor search with HNSW or Faiss to efficiently filter billions of items. The focus then shifts to precision ranking through Deep Interest Networks that capture shifting user preferences. Finally, the course examines re-ranking strategies for multi-objective optimization, balancing immediate engagement with content diversity and long-term retention through reinforcement learning. These briefings provide the necessary framework to build a recommendation engine that balances user intent with platform requirements in a high-scale, real-time mobile environment.
