Leaf-photo crop disease triage that works offline
Smallholder farmers often notice a disease only after it has spread, and the nearest agronomist can be a day away. Rural networks are patchy, so cloud-only diagnosis tools fail exactly when they are needed.
- What to build
- Build a phone app that identifies common crop diseases from a leaf photo using an on-device model, explains treatment options in the farmer's language and queues the case for expert review when connectivity returns.
- Success looks like
- At least 85% top-3 accuracy on a held-out set of locally collected images, results in under 3 seconds with no network, and a feedback loop that flags low-confidence cases.
- On-device ML
- TensorFlow Lite
- Image classification
- Android
- Offline-first sync