// DUAL-STAGE ML INFERENCE
A two-phase classification pipeline: the first model runs lightweight EfficientNetB0 blocks directly on the client device (edge inference), while the second confirmation stage sends a compressed image payload to the cloud API only when confidence falls below threshold — minimizing bandwidth consumption.
// EDGE COMPUTE ARCHITECTURE
TFLite model bundles compiled and shipped inside the React Native application binary. No server round-trip required for standard classifications — results are returned in under 350ms even on 2G rural networks. This design eliminates latency completely for offline-first field use.
// MOBILE-FIRST UX
Designed for farmers and agricultural field workers — not developers. Camera-first interface with large tap targets, offline mode with local result caching, and Malay-English bilingual disease reports with actionable treatment recommendations generated per crop type.