AssetEye by Dronetjek
AI-Powered Drone Inspection Platform for Property Assessment
What is AssetEye?
Dronetjek, a Denmark-based drone inspection company, partnered with Growth Loops Technology to build AssetEye — an advanced, AI-powered platform that transforms the complete property inspection lifecycle. Traditional roof inspections relied on manual work, delayed reporting, inconsistent results, and unsafe rooftop operations. AssetEye automates damage detection across 100+ roof damage types using custom YOLOv5 and RT-DETR models trained on a proprietary dataset of 50,000+ drone images with 200,000+ annotations. The platform integrates AI detection, a Konva.js annotation system, automated severity classification, one-click PDF report generation, and a transparent contractor bidding marketplace — all in one unified workflow. It is the only platform providing AI detection + annotation + reporting + marketplace in one flow.
Manual inspections are slow, unsafe, and inconsistent
Property owners, insurance agents, and contractors faced slow inspection turnaround (3–5 days), unsafe manual roof work, inconsistent damage identification, no standardized reporting, high cost variability, and no transparent contractor bidding. Dronetjek needed a system to detect damage automatically, create annotations quickly, build professional reports, enable repair workflows, and manage contractors transparently.
Dataset Challenges
Dataset imbalance causing inconsistent detection, varying drone altitudes (40ft to 400ft), low-light/shadow issues, multiple roof materials (tile, metal, asphalt, slate), and rare damage types with less than 1% representation.
Workflow Fragmentation
No central platform managing the workflow end-to-end. Manual inspection took 3–5 days with safety risks for on-roof physical inspections and non-standardized reporting.
Pricing Transparency
Lack of pricing transparency in the repair market. No mechanism for competitive contractor bidding or tracking work completion.
AssetEye uses a multi-app modular architecture with AI-driven reporting workflows, real-time WebSocket communication, and a multi-cloud training pipeline.
AI Detection System
Custom YOLOv5 and RT-DETR models detect 100+ damage classes and 10+ roof features. Models handle varying altitudes, lighting conditions, and seasonal drift. Severity mapping classifies damage as Red (critical), Yellow (moderate), Green (minor), or Black (non-damage feature).
Annotation & Reporting Platform
Konva.js-based 2D drawing system supporting polygon, rectangle, and point annotations with real-time collaboration. Automated PDF report generation with images, severity scores, notes, and recommendations in English and Danish.
Contractor Marketplace
Bidding system for transparent contractor engagement. Automated tender-to-contractor-to-work-completion pipeline with job management, enabling the full inspection-to-repair workflow in one platform.
End-to-End Flow
Layered Architecture
Where the hard problems lived
100+ Damage Class Detection
Training models to detect over 100 damage types across multiple roof materials with rare classes having less than 1% representation. Required advanced data augmentation and class balancing strategies.
Multi-Altitude Accuracy
Drones operate between 40ft and 400ft altitude. Models needed to maintain detection accuracy across this range with varying image resolutions and perspectives.
Seasonal Data Drift
Roof appearance changes across seasons (snow, leaves, moss growth). Required continuous retraining and drift detection to maintain model reliability.