Sensor fleets, an MQTT broker, and a time-series store. The backlog was measured in minutes. It needed to be seconds.
2023
Minutes → sub-10 s
Backlog latency
The situation
Smart-building sensor fleets were producing telemetry faster than the ingest path could normalize it. By the time an anomaly was flagged, the building had already moved on.
The work
I built IoT ingestion pipelines: edge-to-cloud normalization and anomaly flagging for the sensor fleet.
I integrated an MQTT broker with a cloud time-series store so the backlog was no longer the product.