AI Key-Recognition System
A computer-vision rebuild that roughly doubled key-recognition accuracy for a property-tech product, from 40% to about 80%.
Our role: Computer-vision module rebuild
The challenge
A property-tech client's key-management product could identify a scanned key only 40% of the time, which made the feature unreliable in real-world use.
What we did
We replaced the underperforming feature extractor with configurable ResNet50 and EfficientNet-B3 models, built a rotation-augmentation pipeline (0/90/180/270°) and image normalization for real-world photos, fixed a vector-database deletion-sync bug, and recalibrated the match threshold to 90%.
The result
Recognition accuracy roughly doubled, from 40% to about 80%, with false positives virtually eliminated. The client is withheld pending permission.
Built with
- Python
- YOLO
- ResNet50
- EfficientNet
- Vector Databases
Client name withheld pending permission.
Have a project like this?
Tell us what you're building. You'll get a written scope and a fixed quote before any commitment.