zoom-BT/ZinEdge

Offline AI diagnostics for health & agriculture in rural Africa, Gemma 3n on-device via LiteRT, zero cloud, zero connectivity required. # ZinEdge — Offline AI Diagnostics for Africa **ZinEdge** is an Android application embedding Google's Gemma 3n model (4B parameters, INT4, 1.9 GB) directly on-device via LiteRT, providing medical and agricultural diagnostics **100% offline** on low-end smartphones. > Built for rural sub-Saharan Africa — no internet, no cloud, no data leaving the device. --- ## The Problem In rural sub-Saharan Africa: - **1 doctor per 10,000 inhabitants** in rural areas - **600M+ people** with no reliable internet access - **Smartphones** cost 80–150 USD (entry-level SoCs: MediaTek Helio, Snapdragon 4xx) Existing AI health tools (GPT APIs, Gemini, Claude) require cloud connectivity. Static decision-tree apps lack natural language understanding. ZinEdge closes this gap. --- ## Features | Module | Description | |---|---| | **Health Diagnosis** | Enter symptoms (fever, chills, cough…) + optional photo → AI produces probable diagnosis, confidence score, and locally-accessible treatment recommendations | | **Crop Disease Detection** | Select crop (maize, tomato, groundnut, sorghum) + photo → AI identifies disease and proposes treatment adapted to locally available inputs | | **Offline Chatbot** | Free-form medical/agricultural Q&A powered by Gemma 3n, fully offline | | **History** | Consult past diagnoses with search and filters | --- ## Technical Stack ``` Model : Gemma 3n E4B-it INT4 (4.92 GB, .litertlm format) Runtime : Google LiteRT (formerly TensorFlow Lite) via MediaPipe LlmInference API Deployment : USB sideload → Android external app-specific storage (no Play Store needed) Platform : Android 8.0+ (API 26+), optimized for NPU (Qualcomm, MediaTek) UI : Jetpack Compose + Material 3 DI : Hilt Local DB : Room Language : Kotlin ``` --- ## Model Deployment (Sideload via USB) The model is **not included** in the APK (4.92 GB exceeds any reasonable app size). Deploy it via ADB: ```bash # Connect phone via USB, enable ADB debugging adb push gem …