AgriHealth AI (UNESCO)
Offline AI for soil & plant disease detection
The challenge
Extension services cannot reach every farm, and the farms that need diagnosis most are exactly the ones with the weakest mobile data coverage. Cloud-only AI tools are unusable in the field.
Offline-first AI that detects plant disease, identifies soil nutrient deficiency and gives farmers advisory guidance without an internet connection.
Our role: AI engineering, mobile development, offline model deployment
Architecture decisions
Inference runs on the device
Vision models are quantised and shipped inside the app, so leaf and soil diagnosis works with the phone in airplane mode.
Two models, one workflow
A plant-disease classifier and a soil deficiency model feed a single advisory engine, so the farmer gets one clear action, not two scores.
Advisory in plain Kiswahili
Every diagnosis maps to locally relevant treatment and nutrient advice written for farmers, not agronomists.
Sync when signal returns
Diagnoses queue locally and upload later, building a regional crop-health picture without ever blocking the farmer.
System architecture
- 01ClientsReact Native app, camera capture, offline UI
- 02AIQuantised PyTorch models exported to ONNX / TFLite
- 03DomainDisease classification, deficiency scoring, advisory rules
- 04DataOn-device SQLite store with deferred sync
- 05OpsVersioned model bundles and over-the-air model updates
Results
Farmers in low-connectivity districts get an instant, understandable diagnosis and treatment plan from a phone camera, and programme teams later receive aggregated crop-health data.
Related ai projects
Planning something similar?
Our Dar es Salaam engineering team builds web platforms, mobile apps, AI features and database systems for clients across Tanzania, Africa and internationally.