ClaimSense
Offline-first motor damage assessment using computer vision and device sensor data.
- Type
- AI / ML
- Client
- Independent insurance technology prototype
- Role
- ML engineering · Mobile systems
- Year
- 2025
Evidence / status
Prototype. Reported outcomes await verification.
Challenge
Assessment at the point of claim
A useful field-assessment tool needed to classify visible vehicle damage quickly without assuming a reliable network connection or dedicated mobile GPU.
Constraints
- Core inference had to work offline.
- Per-frame latency needed to support live camera feedback.
- The model needed to distinguish multiple damage categories.
- Reports needed to sync safely after connectivity returned.
Approach
Key decisions
Single-pass damage detection
A fine-tuned YOLOv8 model detects and classifies several visible damage categories in one inference pass.
ONNX for device inference
The trained model is exported and quantized for a smaller runtime footprint on Android hardware.
Independent sensor-risk signal
Accelerometer and gyroscope readings create a second risk input instead of overloading the vision model.
Offline-first report storage
Assessments are written locally first and synchronized only when connectivity is available.
Evidence
What can be stated
Current status: Prototype. Unverified outcomes are labelled instead of presented as proof.
- Review source (opens in a new tab)
The repository notes report 94.2% mAP and sub-100ms inference on a test device.
Needs verification
Technology stack
- Python
- YOLOv8
- ONNX
- Firebase
Next project
Continue through the work archive.
Start a project
Let's build something great
Drop your details below. We'll have it open in Gmail, ready to send — no copy-pasting.