NYC Taxi Supply/Demand Modeling

Design a real-time system that models and predicts taxi supply and demand across the 260 official TLC zones of New York City. The system must ingest live trip-start events, weather feeds, scheduled events (flights, games, concerts), and road-closure data; within a 5-minute window it must emit for every zone (1) predicted ride-request count (demand), (2) predicted number of available taxis (supply), and (3) a recommended surge multiplier and driver-repositioning bonus. The service will be called every minute by the pricing engine and the driver-app heat-map service (≈ 50 k QPS combined). Your design should handle 1 M trip starts/hour at peak, 300 k active drivers, and 5-year historical data (≈ 3 TB). You must explain how you partition the problem into (a) demand forecasting, (b) supply forecasting, (c) mismatch computation, and (d) incentive generation; justify your choice of spatial-temporal models, feature stores, streaming platform, and serving layer; discuss cold-start for new zones, model-update cadence, back-testing, and A/B infrastructure; and provide an end-to-end latency & availability SLA with fallback strategy.