Jev Telecom Alarm Triage Demo

Turning an alarm flood into an operator queue A trunk fiber cut can trigger errors across many downstream devices. A utility outage can put several cell sites on battery at once. This telecom NOC demonstration shows a proposed triage workflow: code assembles topology, readings, history and change records; a decision layer makes bounded judgments; explicit policy determines the operator queue. The video uses an English interface and has no audio. Its opening states that Jev answers are scripted, while the policy, priority floors and timers run in the demonstration. It illustrates a workflow rather than a live network deployment or a model evaluation. Five scenarios - Trunk fiber cut: Group downstream alarms by their shared cause using topology, while preserving priority floors for critical hospital and bank services. Related alarms remain traceable, and a nearby unrelated alarm stays separate. - Nine cell sites lose utility power: Use remaining battery runtime to organize attention. Low battery and unacknowledged incidents trigger policy escalation. An accelerated simulated clock illustrates the timers. - Night maintenance window: Summarize expected restarts and link changes, but surface out-of-scope faults and sites that remain down after the window closes. - Emergency calling incident: For a simulated emergency-call setup rate of 91%, inject an incorrect “routine, 0.72” answer. A downgrade gate and priority floor block the downgrade; an independent timer later escalates the unacknowledged incident. - Jev outage: Fall back to the device vendor’s original severity and retain the alarms, demonstrating the path used when no model answer is available. Where the decision model fits The interface separates six judgments: severity, responsible team, match to historical patterns, shared cause with an existing incident, whether an alarm can wait, and impact. Ordinary code handles facts, thresholds, downgrade conditions, priority floors for critical services, timer escalation and the audit trail. The model answers, latency figures and token counts in this recording are demonstration data. They do not establish real Jev accuracy, response time or cost. The recording also does not validate missed-alarm rates, false alarms or response performance on a live telecom network; those require real incidents, human labels and operational testing.