Detect maize disease
before it spreads.
MaizAI puts a trained AI in the farmer's pocket. Snap a leaf and get a diagnosis in seconds, even without an internet connection. Designed for the realities of smallholder farming in Cameroon.
Trained on the PlantVillage maize dataset · Validated on Cameroonian field conditions
Under 3 seconds
Leaf to diagnosis, offline
Late diagnosis costs Cameroonian farmers up to 35% of their maize harvest.
* Frontiers in Plant Science, 2026. Maize disease and pest detection in sub-Saharan Africa.
Disease spreads fast
Common Rust, Gray Leaf Spot, and Blight can devastate a plot within days under warm, humid conditions.
Connectivity is limited
Many farms lack reliable internet access, making cloud-based diagnostic tools impractical on the plot.
Early action saves crops
A diagnosis in the first 48 hours gives farmers time to apply targeted treatment before spread is uncontrollable.
How it works
From leaf snap to recommendation in under 3 seconds.
Capture the leaf
Open MaizAI on your Android phone and photograph the affected maize leaf.
On-device AI classifies
The TFLite model runs locally, so no internet is required. Four classes: Healthy, Common Rust, Gray Leaf Spot, Blight.
Sensor data adds context
The ESP32 node provides live soil moisture, temperature, and humidity to refine the urgency level.
Recommendation issued
The rule engine issues a targeted intervention (spray now, monitor, irrigate, or remove) matched to field conditions.
Environmental context, direct from the field.
The custom ESP32 node sits in the soil between maize rows, continuously monitoring soil moisture, ambient temperature, and relative humidity. It broadcasts live readings locally when the farmer is on plot, and syncs to the cloud via MQTT when they are away, so the rule engine always has context.
- ▸Soil moisture · Temperature · Humidity
- ▸mDNS local discovery, no cloud needed on-plot
- ▸MQTT → HiveMQ Cloud when farmer is away
- ▸Low-power, field-deployable design
Built for the field
Not a lab tool. A field tool.
Offline-first
The AI lives in the app. No signal needed to diagnose a leaf. Data syncs to the cloud when connectivity is available.
Context-aware
Pairs disease detection with live soil moisture, ambient temperature, and relative humidity from the ESP32 sensor node.
Works on modest devices
Designed for Android phones with limited RAM. The quantised TFLite model runs in under 1 second on a mid-range device.
Three-layer architecture
Mobile, cloud, and sensor, each doing what it does best.
Mobile (React Native)
- ▸TFLite disease classifier (on-device)
- ▸mDNS sensor discovery
- ▸On-device rule engine
- ▸Offline-first data sync
Cloud (Next.js + Neon)
- ▸REST API for sync & query
- ▸Admin dashboard
- ▸Cloud rule engine
- ▸Cloudinary image storage
Sensor (ESP32)
- ▸Soil moisture, temp, humidity
- ▸mDNS local advertising
- ▸MQTT → HiveMQ Cloud
- ▸Low power, field-deployable
Get MaizAI on your phone
Free, open source, no account required to start diagnosing.