AI-PoweredCatastrophe Models

Fast, global, location-level models of hail, tornado and straight-line wind, built and validated for insurance decisions.

U.S. National Science Foundation sealU.S. National Science FoundationNSF SBIR Phase I & II
Trusted by industry leaders
Britt InsuranceLloyd's LabAonEmergentUC Berkeley
Britt InsuranceLloyd's LabAonEmergentUC Berkeley

WHY AI

No single source
sees the whole storm.

Traditional cat modellingRadar, atmospheric data and storm reports describe different parts of a storm. Bringing them together through hand-written physics and heuristics means coding relationships, exceptions and approximations.

Vāyuh’s physics-informed AILearn the relationships across these sources, reducing the need for hand-written rules and enabling us to generate more storm scenarios, faster.

Explore our storm models
Physics-informed AI
Learn across sources. Simulate at scale.

SEVERE CONVECTIVE STORM MODELS

Built around each peril.

Hail, tornado and straight-line wind behave differently. Our models reflect those differences, with flexible event definitions for how you assess risk.

  • Hail

    Hail size, frequency and local footprints, from individual storms to large outbreaks.

  • Tornado

    Tornado tracks, widths and intensity, resolved around the locations that matter to you.

  • Straight-line wind

    Wind speed and damaging footprints across storm types, from downbursts to derechos.

  • Hazard maps
  • Event catalogs
  • Loss estimates

Coverage, resolution and delivery vary by peril and region.

THE STORM CATALOG

Better loss estimates.
Built on millions of simulated storms.

Large simulated storm catalogs reduce sampling noise, giving you a more stable estimate of loss at each location.

Explore portfolio risk
ONE LOCATION · SAME MODEL

A more stable view of loss.

Sensitive to a few stormsLess sampling noiseFewer yearsMore years

Estimated annual loss

Simulated years

LOCATION-LEVEL HAZARD

Detailed risk at each location,
calibrated to extremes.

Local hail intensity, checked against historical extremes.

Review calibration and evidence
9 MAY 2024 Texas → Southeast hail outbreak
TEXASOKLAHOMALOUISIANAGEORGIATENNESSEE123 km modelledResolution
ReportsHail ≥ 2 inUnder-predicted
Local aerial imagery: USGS / USDA NAIP

At the extremes.

LocationObservedVayuh
1Texas6.12 in6.00 in
2Alabama4.50 in4.18 in

Hail size in inches.

98%

183 of 187 reports had severe predicted hail within 50 km.

ASSET-LEVEL INTELLIGENCE

Asset details
shape the risk.

Vehicles in an open lot, covered inventory and building roofs respond differently to the same storm. Understand what is exposed at each site.

Illustrative aerial view of a dealer open lot with exposed vehicles, cars beneath a canopy and a separate showroom roof
Covered inventoryUnder a canopy
Building roofFlat / low-slope
Exposed inventoryVehicles outdoors

LOSS MODELLING

An alternative
view of risk.

For insurers, reinsurers and brokers comparing location-level hazard and portfolio loss alongside existing models.

Hazard data
Event footprints, storm frequency and intensity at your locations.
Loss estimates
Average annual loss and loss-cost comparisons, based on exposure and vulnerability assumptions.
Delivery
Event sets for the Oasis Loss Modelling Framework (Oasis LMF), or a portfolio analysis prepared with our team.

GLOBAL SEVERE CONVECTIVE STORM MODELS

Global storm
risk models.

A consistent view of severe convective storm risk for portfolios that cross markets.

CASE STUDIES

See the models
in action.

Focused studies of storm risk across different assets.

THE STANDARD

Insurance-grade
risk intelligence.

Evaluate the assumptions, the evidence and what the results mean for your portfolio.

Run it on your locations.

  • Hazard

    Modelled footprintObservations

    Observed storms.

    Review frequency and intensity against observations.

  • Event sets

    Same locationMore events · less sampling noise

    Local stability.

    Examine storm footprints and how estimates converge.

  • Calibration

    VulnerabilityExposureTrace the inputs behind the estimate

    Tailored to you.

    Calibrate to your assets and experience, with adjustments documented.

INDUSTRY PERSPECTIVES

In their words.

Quote 01 / 03

Vayuh's hail model is the best I've seen in my 15 years working on severe convective storms. Their approach is fundamentally different and superior to anything else in the market. This is the future of risk modeling.

Dave VicaryRenew RiskCustomer

Company

Built by scientistswho ship.

A team out of the UC Berkeley AI/ML research community — supported by the U.S. National Science Foundation under SBIR Phase I & II, and backed by operators across insurance, infrastructure, and ML systems.

START WITH YOUR PORTFOLIO

See what the models
mean for you.

Choose a peril and a few locations. We’ll scope the outputs and review the results, assumptions and available evidence together.

Run it on your locations.