ML System Design (Recommendation / Feed Ranking / ETA)

This is an open-ended system design problem for Uber. Requirements Functional: State precisely what the model should predict: click-through rate, conversion, ETA error, or matching success. Lay out the feature pipeline across offline batch jobs, online real-time streams, and entity-level joins. Pick from model families such as gradient-boosted machines, wide-and-deep, two-tower retrieval, or transformer rankers. Deliver predictions under a p99 latency cap — commonly under…