Eugene Yan's article "Real-Time Machine Learning For Recommendations" examines the growing shift from batch processing to real-time ML in recommender systems. It asks why real-time matters, how companies in China and the US have built such systems, and how teams should approach an MVP.
The article argues that real-time recommendations allow systems to incorporate the latest user signals, making suggestions more relevant and responsive. It notes that companies in both regions have invested in real-time infrastructure, though their designs differ depending on scale, product context, and engineering constraints.
For teams getting started, the article recommends keeping the first version small: choose a specific use case, avoid overbuilding, and iterate from there. The emphasis is on solving a concrete problem first rather than constructing a full-scale real-time platform on day one.