The Challenge
What MaxSense Was Facing
MaxSense monitors industrial equipment across manufacturing sites using thousands of IoT sensors. The engineering problem was ingesting a continuous stream of 50M+ sensor readings per day, detecting anomalies in near-real-time, and storing data efficiently for 3-year retention without the storage costs becoming prohibitive. Existing time-series databases had been unable to handle the write throughput without degrading query performance.
The Solution
What We Built
We designed the ingestion pipeline around AWS IoT Core as the MQTT broker, feeding into Kinesis Data Streams with per-sensor sharding for ordered delivery. Consumers wrote to InfluxDB clusters with a tiered retention policy: raw data for 30 days, downsampled 1-minute aggregates for 1 year, and hourly aggregates for 3 years. An anomaly detection service consumed the Kinesis stream in parallel and evaluated sensor readings against ML-trained baseline models, publishing alerts to SNS within 90 seconds of anomaly onset. Infrastructure scaling was handled by ECS Fargate with Kinesis Consumer Library for automatic shard rebalancing.
Results