Inspection & damage
Defects found, classified and graded, on vehicles, buildings and infrastructure.
Damage and defect annotation for inspection models and the insurers who rely on them, with live review of low confidence detections.
From the first photo to the settled claim
Every part of the workflow where we can take work off your team.
Capture
Assessment
Claims
Fraud and audit
We work with photos, drone imagery, thermal and underbody.
What we label
Every class is written into a guide with your team before the first batch. Files come back as COCO JSON, YOLO, Pascal VOC, CSV, JSONL or your own schema.
Vehicles
Parts
Polygon and bounding box
Bumper, door panel, headlight, mirror, wheel
Damage
Polygon and mask
Dent, scratch, crack, rust, broken glass
Severity
Class
Minor, moderate, major, repair or replace
Plates and VINs
OCR
Plate characters, visibility, VIN
Buildings and infrastructure
Facade and roof defects
Polygon and severity
Crack, spalling, missing tile, water ingress
Thermal anomalies
Bounding box and class
Heat loss, moisture, electrical hotspot
Linear assets
Polyline and bounding box
Pipeline, transmission line, bridge joint
Claim documents
Forms and policies
OCR and key value
Policy number, coverage limit, deductible, incident date
How we test the model
The tests that matter for inspection & damage models, and what you get back from each.
Severity agreement
Model severity grades compared with grades from trained reviewers.
You get backAgreement by part and damage type
Missed damage audit
Reviewers look for damage the model did not flag.
You get backMiss rate by part, examples for retraining
Fraud flag review
Flags on manipulated or reused photos checked by a person.
You get backFalse positive rate on fraud flags
Where inspection & damage teams start
Problems teams bring us, and the first piece of work we would scope.
Your model finds the dent but grades it wrong, and the payout follows the grade.
What we doSeverity labels with a written scale, and disputed grades reviewed twice.AnnotateClaim photos arrive in every light, angle and phone camera.
What we doWe collect and label across conditions, so the model meets real submissions in training.CollectLow confidence assessments wait in a queue for an adjuster.
What we doOur reviewers check them live and return a decision with a reason.Live reviewRelated work
200,000 polygons for automated vehicle damage assessment
Defect annotation on building imagery in CVAT, with a webhook loop between annotators and reviewers.
50,000 facade annotations a week for building condition assessments
2,000 robot videos described every week by a team built in seven days
Half a million images for autonomous navigation, with every file kept on EU servers
Bring us the inspection & damage data your model gets wrong.
Send a small batch and get it back done, free, usually within 24 to 48 hours. You judge the work before you pay for anything.