Problem
Design an ML system that predicts real-time remaining trip ETAs updating every few seconds during active rides
- Correct GPS noise and map-match coordinates to road segments before any feature lookup or label construction
- Predict per-segment travel times under 100ms while bounding prediction jitter to maintain rider trust
- Handle sparse fleet telemetry and mid-trip route deviations without serving stale cached values
Key Test Points
- HMM-based map-matching
- Segment decomposition
- XGBoost regression
Spot an issue or have a suggestion?