01The challenge
The utility's smart-meter rollout was outrunning its data platform. The incumbent vendor quoted a seven-figure annual bill for an ingestion tier that still dropped readings during storms — exactly when the grid team needed them most. Anomaly reports arrived a day late, as batch jobs.
02The approach
We replaced the batch pipeline with a streaming architecture: Kafka ingestion sized for storm-day peaks, time-series storage tiered by access pattern, and anomaly detection that scores readings in-stream instead of overnight. Every component is boring, proven, and sized from measured load — not the vendor's sizing guide.
DROPPED READS · DAY-LATE ALERTS
03The result
Grid operators now see anomalies in under a minute, storm-day ingestion holds at 99.98% completeness, and the platform runs on infrastructure the utility's own team can operate.