Project background
Electrical operators collect enormous telemetry streams but rarely act on them in real time. The client wanted a platform that surfaced actionable fault signals as they emerged, not hours later in a report.
Challenge
Keeping ingest, analytics, and alerting latency low enough for operational use while scaling across hundreds of feeders. Tuning anomaly detection to minimize false positives without missing real events.
Approach & solution
We built a streaming analytics pipeline on a purpose-built time-series store, with anomaly detectors running per feeder and aggregating to substation-level views. Alert thresholds are learned per asset rather than fixed. Operators see both live and historical context for every alert.
Results & benefits
Operations teams now investigate and resolve events in real time rather than after the fact. Alert volume decreased once thresholds were tuned per asset, and missed-event rates dropped in parallel.
