The challenge
The monitoring in place was the vendor default: each sensor reporting to a basic cloud dashboard that showed current values on a public page — and nothing else. No historical trending to catch a pump cycling more often than it should. No anomaly detection to flag a tank level drifting out of range. No alerts to anyone's phone. For infrastructure where a missed failure means an environmental incident and an emergency call-out, "glance at a webpage" isn't monitoring.
What we built
We redirected the entire fleet — 17 IoT sensors — through a self-hosted telemetry pipeline:
- MQTT broker as the single ingestion point for every sensor on the property.
- Time-series database recording every reading, with retention limited only by disk — not by a vendor's plan tier.
- Grafana dashboards purpose-built for the domain: pump cycle counts and durations, tank levels against thresholds, and component-health panels per site.
- Push notifications the moment a reading crosses an anomaly threshold — a pump running long, a level rising too fast — sent straight to phones.
The cutover ran sensor-by-sensor with the old feed left running in parallel, so there was never a monitoring gap. Total migration time: one week.
How it works
Sensors publish over MQTT to the self-hosted broker; a collector writes every message into the time-series database; Grafana reads from it for live dashboards and evaluates alert rules continuously, pushing notifications on breach. Every component is open-source and runs on infrastructure the operator controls.
The results
The operator went from raw numbers on a public page to real-time anomaly alerts, unlimited history for every one of the 17 sensors, and dashboards that show component health at a glance — pump behavior trends that were invisible before now surface days before a failure would. Ongoing platform care is a Run engagement: we keep the pipeline healthy, patched, and alerting.
Where this pattern fits
Any operation with field sensors trapped in vendor dashboards — septic and water systems, HVAC fleets, cold storage, agriculture — can run this exact pipeline. The stack is open-source, the data lands somewhere you own, and "monitoring" starts meaning someone gets woken up before the failure, not the graph existed.