home / case studies / Indian electric mobility OEM (composite)
Data Streaming

Streaming Telemetry Platform for a Connected Two-Wheeler Fleet

We built a streaming pipeline that ingests battery, GPS and motor telemetry from a growing fleet of connected vehicles. Operations teams now get live alerts and battery-health insights instead of weekly CSV exports.

Indian electric mobility OEM (composite) · Data Streaming

Streaming Telemetry Platform for a Connected Two-Wheeler Fleet

Challenge

Patchy rural connectivity meant bursts of delayed, duplicated and out-of-order messages. We added device-side sequence numbers, idempotent sinks and late-data side outputs so analytics stayed correct without dropping data.

Approach

Vehicles publish MQTT messages to an EMQX broker cluster bridged into Kafka. Flink jobs run windowed aggregations (state-of-charge trends, thermal anomalies, trip segmentation) with event-time processing and watermarks to handle out-of-order mobile network data. Hot data lands in a time-series store (TimescaleDB) for dashboards; cold data goes to Parquet on object storage for data science. A rules service lets operations staff define alert thresholds without code deployments.

Outcome

Thermal and battery anomaly alerts in near real time instead of after depot sync Illustratively ~30–50% faster warranty-claim root-cause analysis Platform scaled horizontally as the fleet grew, without re-architecture Data science teams got clean, partitioned historical data for model training Operations can tune alert rules without engineering tickets

✓Thermal and battery anomaly alerts in near real time instead of after depot sync
~30–50%Illustratively faster warranty-claim root-cause analysis
✓Platform scaled horizontally as the fleet grew, without re-architecture

Problem solved

Telemetry was uploaded in batches when vehicles reached depots. Battery faults and theft events were found long after the fact, and warranty analysis took weeks of manual data wrangling.

What we built

Vehicles publish MQTT messages to an EMQX broker cluster bridged into Kafka. Flink jobs run windowed aggregations (state-of-charge trends, thermal anomalies, trip segmentation) with event-time processing and watermarks to handle out-of-order mobile network data. Hot data lands in a time-series store (TimescaleDB) for dashboards; cold data goes to Parquet on object storage for data science. A rules service lets operations staff define alert thresholds without code deployments.

Benefits

Thermal and battery anomaly alerts in near real time instead of after depot sync Illustratively ~30–50% faster warranty-claim root-cause analysis Platform scaled horizontally as the fleet grew, without re-architecture Data science teams got clean, partitioned historical data for model training Operations can tune alert rules without engineering tickets

MQTTEMQXKafkaApache FlinkTimescaleDBParquet

Talk to us