OnPoint Pulse
AI-Attended Network Operations

Always-On Power, Always Watched.

Pulse connects your standby generators directly to an AI-attended operations center — live telemetry, deterministic alert engine, AI triage, service history, and dispatch coordination in a single closed loop. When Generator 7 fails to start at 02:14, we know in 60 seconds, your facility manager knows in 5 minutes, and a tech has context before the truck leaves the yard.

GensetCellularCloudAI-NOC
Safety-adjacent infrastructureFail loud. Never fail silent.

The moat

Monitoring hardware is commodity.
The closed loop is not.

Competitors like Power Telematics are fundamentally dumb pipes: device to cloud to SMS blast with zero context. Our moat is the closed loop between live telemetry, AI diagnosis, service history, and dispatch — and nobody selling monitors owns the service relationship. We do.

Dumb pipe (competitors)

  • Device → cloud → SMS blast to every contact on file.
  • Zero context. The message says "Generator fault" — no severity, no history, no diagnosis.
  • Comms failure looks identical to generator failure. Silent device = false confidence.
  • Every alert lands on a human. Every human makes the same uninformed judgment call.
  • No closure loop. Dispatch outcome never feeds back into monitoring.

OnPoint Pulse

  • Telemetry → deterministic alert engine → AI triage in < 60 s.
  • Every ticket arrives pre-classified: generator vs. telemetry, severity P1–P4, ranked diagnosis, service history highlights.
  • Comms layer is monitored. "Device silent" triggers its own remediation ladder before any alarm fires.
  • 70%+ of alerts resolved or correctly classified without dispatch. Humans only see tickets that need them.
  • FSR outcome feeds back to the AI. Diagnosis accuracy compounds over every service call.

How it works

From genset to classified ticket in four steps.

Telemetry at the source

LTE telematics gateways speak Modbus, J1939, and CAN to the generator controller. Every reading — frequency, voltage, oil pressure, battery voltage, fault codes, run state — publishes over MQTT/TLS every 60 seconds at rest, immediately on alarm. 72-hour store-and-forward means a comms gap never loses data.

Deterministic alert engine

Cloud ingestion normalises every controller's readings into one canonical schema. A deterministic rule engine — with hysteresis, debounce, and schedule awareness — fires clean, well-formed events. No AI guesses involved at this layer. Heartbeat watchdog, exercise calendar, and dying-gasp correlation run here.

AI-NOC triage in < 60 s

Every event enters an Inngest workflow. Stage 1 answers the first question: generator fault or telemetry fault? Stage 2 pulls the asset's full OfficeHub service history, queries Rudy's generator knowledge base for ranked root causes, and writes a classified ticket: severity P1–P4, diagnosis hypothesis, recommended action, draft customer message. All in under 60 seconds.

Customer comms and dispatch

P1 failures trigger a customer notification within 5 minutes — plain English, specific facts, no alarm-blast. Where the AI recommends dispatch, a draft work order lands in OfficeHub with the diagnosis attached. FSR outcomes feed back into the AI so diagnosis accuracy compounds with every service call.

Capabilities

Everything the dumb pipe left out.

AI triage and diagnosis

Every alert is classified generator vs. telemetry before any human is looped in. The AI ranks root causes by pulling Rudy's generator knowledge base, the asset's service history, and fleet-wide fault patterns. P1 failures land on a coordinator's desk pre-diagnosed — not as raw alarm codes.

Comms self-healing

A silent device is not automatically a broken generator. Pulse runs a remediation ladder — SIM signal check, carrier failover attempt, remote device reboot — before escalating to an alarm. Over 80% of comms-layer issues resolve without waking anyone up.

Honest monitoring status

No telemetry surface shows last-known state as if it were current. Every reading carries a freshness timestamp. If the device has been silent for 18 minutes, the dashboard says so — explicitly, in the UI, not buried in metadata. Stale data is labeled.

Exercise and NFPA-110 readiness

Exercise calendars are per-asset. Missed, failed, or short exercise runs are detected within 24 hours of the scheduled window. Battery voltage trending, fuel level, and last-run duration feed a readiness score the customer portal displays in plain English.

OfficeHub work-order loop

When the AI recommends dispatch, a draft work order arrives in OfficeHub with the diagnosis, fault codes, and recommended parts attached. FSR completion closes the loop — outcome data feeds back to the AI and trains future diagnosis. The service relationship is the moat.

Fleet analytics

BigQuery holds every normalized telemetry reading, event, and AI decision — partitioned by day, clustered by device. Battery voltage decay curves, controller-family fault rates, exercise pass rates, and mean time between failures are queryable across the entire fleet.

Product Demo

See exactly what we built.

No stock screenshots. No marketing mockups. These are the actual product screens — telemetry values, AI reasoning trails, and autonomy policy tables from the live platform.

Live generator status across all sites

Fleet Overview

4 sites · 6 generators

Live

GEN-A1

Hospital — Bldg A

Online

Standby Ready

Batt 12.7 VFuel 82%
Live

GEN-A2

Hospital — Bldg B

Online

Standby Ready

Batt 12.5 VFuel 76%
Live

GEN-B1

Telecom Tower 14

Degraded

Standby Ready

Batt 11.9 VFuel 61%
2 h stale

GEN-C1

Data Center East

Fault

Fault: FTS

Batt 12.4 VFuel 88%
Live

GEN-C2

Data Center East

Online

Running

Batt 12.8 VFuel 91%
Live

GEN-D1

Warehouse Complex

Offline

Unknown

Batt — VFuel
Offline since Jun 13

By the numbers

Design targets, not marketing claims.

These are the product success metrics from the Pulse specification — targets we build to and measure against, not retroactively applied marketing language.

< 60 s
Raw alert to classified ticket
< 5 min
P1 customer notification, 24/7
99.9%
Alert-path uptime
> 70%
Alerts resolved without dispatch
100%
Missed exercise detection within 24 h
< 5%
False-dispatch rate target

Platform doctrine

Fail loud. Fail toward notification. Never fail silent.

A missed fail-to-start alert during an outage is the worst failure mode in this system. When the pipeline degrades — AI timeout, network partition, model unavailability — the fallback path fires a direct severity-mapped notification without waiting. The dumb path is not a fallback of last resort. It is the most important code in the repo, and it has a dedicated end-to-end test that runs on every pull request.

Always On.Always Watched.Always Answered.

Get started

Every OnPoint service customer deserves to know their generator is ready before the lights go out.

Pulse is in active deployment. Request a demo walkthrough — live telemetry pipeline, AI-NOC triage, and the customer readiness portal — with a real generator fleet.

An OnPoint Generators platform  ·  Safety-adjacent infrastructure  ·  Fail loud. Never fail silent.