Palantir System Design: Taxi Route Recommendation Service

You are asked to design a backend system that helps taxi drivers optimize daily routes using ~8,000 historical routes and live signals. The system should produce near-real-time, personalized route recommendations that maximize earnings or minimize time, and integrate with existing taxi apps via an API.

Core problem: ingest historical and live telemetry (GPS traces, trip durations, fares, traffic, events), process it into features, and serve low-latency recommendations tailored by driver location, time of day, and demand patterns. You must balance scalability, freshness, high availability, and cost.

High-level flow:

• Data ingestion: batch import of historical routes + continuous ingestion of driver telemetry and traffic feeds (streaming).
• Processing & features: stream processors (Kafka + Flink/Beam) for near-real-time features; nightly/periodic batch jobs for heavy aggregations and model training.
• Recommendation engine: hybrid approach — fast heuristic / rules + light-weight ML ranking model for personalization; serve via a low-latency model server behind a cache and API gateway.
• Feedback loop: collect driver accept/decline, ratings, and actual trip outcomes for online or periodic model updates.

Skill signals interviewers look for: distributed systems design (message buses, sharding, partitioning), stream processing and stateful operators, feature engineering for time-series and geospatial data, online vs offline model training, API design for mobile integration, latency/availability trade-offs, and cost controls (sampling, TTLs, spot instances). You should explain failure modes, monitoring, and how feedback improves recommendations over time.