VERMILION · LRM · IN PRODUCTION

Predict the unpredictable.
Run undisturbed.

Vermilion is a physics-informed Large Reasoning Model that anticipates equipment failure days before it surfaces — and explains its reasoning in your engineers' language.

Reasoned
Informed physics
Auditable
Every prediction,
every stage
Eight
Industrial sectors served today
ANOMALYVIBRATION · ETF 96h
STREAM · OK600 Hz
FLEET · MRO · FAN-3
PHYSICS-INFORMED · LRMFAILURE WINDOW · 96h ± 11h
HOW IT WORKS

Reasoning, informed physics pattern matching

Most predictive maintenance is curve-fitting in disguise. Vermilion combines physical first principles with reinforcement-tuned reasoning so you get answers your reliability team can actually act on, and verify.

01 / 03

Sense

Ingest sensor telemetry, maintenance logs and equipment specs from any source: historians, edge gateways, CSV exports. No protocol lock-in.

02 / 03

Reason

The LRM grounds its inferences in physical laws — conservation of energy, Newtonian dynamics, fatigue mechanics — refusing physically impossible conclusions.

03 / 03

Recommend

Get specific, time-bounded actions: which part, what window, why now. Every recommendation comes with the reasoning chain attached.

REASONING TRACE · LIVE

Watch Vermilion think through a bearing failure.

Every prediction is a chain of physical inferences, not a black-box score. Pause at any step. Inspect the evidence. Override and re-run.

Grounded in physics
Inferences must satisfy known physical laws; never produces an answer that violates conservation of energy.
Calibrated uncertainty
Confidence is a probability, not a vibe. Calibrated against thousands of real outcomes.
Cross-asset learning
Patterns discovered on a paper-mill bearing in Finland improve predictions on an aerospace gearbox in Quebec — privately, with strict data isolation.
Read the technology brief
REASONING TRACElive
1
INGEST
Streaming sensor telemetry from Bearing BRG-04 · vibration, temperature, current draw.
2
PHYSICS
Vibration spectrum shows a resonance peak at 120 Hz. Inconsistent with healthy bearing harmonics.
3
REASON
Pattern matches inner-race wear. Conservation-of-energy bound on heat dissipation rules out outer-race fault.
4
PROJECT
Damage progression model projects failure at 96 ± 11 hours, p = 0.93. Rising amplitude consistent with surface spalling.
5
ACT
Recommend swap during planned downtime Friday 02:00–04:00. Part: SKF-23140. Avoids unplanned 14h outage.
SOURCE: 1 ASSET · 4 SIGNALS · 312k SAMPLESCONFIDENCE 0.93
CAPABILITIES

One platform.
Limitless applications.

A single reasoning model adapted to your assets: pumps, presses, turbines, gearboxes, lines. Cross-domain learning lifts every deployment as the fleet grows.

01 / 06

Physics-informed core

Neural networks constrained by Newton, thermodynamics and fatigue mechanics, so the model refuses to hallucinate impossible answers.

02 / 06

Cross-industry insight

Solutions discovered on one fleet privately improve predictions on another. Anonymized, isolated, opt-in.

03 / 06

Virtual maintenance expert

Conversational interface for your reliability team. Ask why. Get reasoning. Override and re-run.

04 / 06

Edge-to-cloud architecture

Deploys at the edge for latency-sensitive sites and in our cloud for fleet-wide reasoning. Air-gapped on request.

05 / 06

Calibrated uncertainty

Probabilities you can trust at the planning desk. We publish reliability diagrams every quarter.

06 / 06

Open integration

REST, gRPC, OPC-UA, MQTT, Historian. Bring your own data lake. Bring your own SCADA.

INDUSTRIES

Wherever uptime decides outcomes.

Our platform adapts to any industry where equipment reliability is critical. Physics doesn't change between sectors. Neither does Vermilion.

OUTCOMES

The signal beneath the noise.

Across our deployments, Vermilion catches what others miss. These are the numbers we keep on the wall.

96h
Median advance warning before failure
93%
Of predictions verified within ±15% of projected window
Reduction in false-positive maintenance alerts vs. legacy systems
100%
Of predictions ship with a 5-stage reasoning trace and calibrated confidence
ILLUSTRATED SCENARIO

A silent bearing fault, caught four days early.

Vermilion flagged a 120 Hz resonance on a primary crusher bearing. SCADA showed nominal vibration; the physics model didn't agree. Replacement during scheduled maintenance avoided 14 hours of unplanned downtime.

14 h
Unplanned downtime avoided
96 h
Lead time on the prediction
1 part
SKF-23140 swap during planned window
BRG-04 · PRIMARY CRUSHER
ACTION REQUIRED
VIBRATION SPECTRUM · 0–500 Hz
120 Hz · INNER RACE0125250375500 Hz
FAILURE WINDOW
96 ± 11 h
CONFIDENCE
0.93
Recommendation. Replace bearing during planned downtime Friday 02:00–04:00. Part: SKF-23140. Estimated repair: 2 h. Avoids ≈ 14 h unplanned outage.
QUESTIONS

What enterprise buyers ask first.

Most predictive-maintenance systems learn statistical correlations from historical data. Vermilion grounds its reasoning in physical laws — conservation of energy, fatigue mechanics, fluid dynamics — so it generalizes to assets and conditions it hasn't seen, and refuses to make predictions that violate physics.
A pilot on a single asset class typically goes live in 4–6 weeks: two for data ingestion and physics-model fitting, two for shadow operation against your existing alerts, and one to two for handoff to your reliability team.
Your data is isolated in a tenant-only environment. Anonymized signal patterns can, at your option, contribute to and benefit from cross-fleet learning.
Both. Latency-sensitive asset classes run an edge LRM colocated with the historian; fleet-wide reasoning runs in our cloud, or yours, or air-gapped. We support OPC-UA, MQTT, REST and gRPC.
Three things: catastrophic unplanned failures trending toward zero, false-positive alerts cut by an order of magnitude, and your maintenance planners using the reasoning trace as a daily tool, not a dashboard they ignore.
READY WHEN YOU ARE

Move from reactive
to predictive.

See how our reasoning core thinks about a failure mode that matters to your operation. Thirty minutes, with one of our reliability engineers.