FoodAR
A mobile app that detects mold on food in real time with a custom YOLO AI model, at 91% accuracy.
- Role
- Design, AI model training and development
- Built with
- Flutter, Python, YOLO AI, TensorFlow Lite, Roboflow, Firebase, Figma
The brief
At first glance it's hard to tell harmless discoloration from toxic mold, and getting it wrong can mean an allergic reaction or contact with harmful spores. The app had to recognise mold in the camera image and warn the user clearly.
What I did
I trained a custom YOLO model to detect mold on food, using Roboflow and Python. The model runs inside the app with TensorFlow Lite, so the camera image is analysed live. Firebase stores the app's data.
I designed the interface in Figma and built it in Flutter. The app shows the result on the camera preview and warns about possible risks such as allergies or toxic spores. The “How To?” onboarding screen teaches people to take a photo the model can read: not messy, not too close and not through glass.
Key features
- Real time mold detection with a YOLO model
- 91% accuracy telling mold apart from discoloration
- Warnings about health risks
- Onboarding that teaches people to take readable photos
Result
The model tells harmless discoloration apart from toxic mold with 91% accuracy.
Have a similar project in mind?
Next projectKombuczara