100% Offline · AI-Powered · Android & iOS · 6 Regional Languages
Over 65% of Indian farmers operate in low-connectivity rural areas. When crops fall sick or machinery fails, immediate expert help is unavailable.
This leads to:
- Delayed treatment
- Crop yield loss
- Financial stress
- Dependence on middlemen
Most agricultural AI solutions require internet — making them unusable where they are needed most.
Farmphile AI solves this gap by bringing 100% offline AI directly to farmers' smartphones.
Farmphile AI is a hackathon submission that puts the power of AI directly in the hands of Indian farmers — no internet required, ever. Built with Expo React Native and powered by the AnyWhere SDK for fully on-device neural network inference, farmphile AI runs on any Android or iOS smartphone.
Farmers can photograph their crops, record machinery sounds, and access a curated knowledge base — all without a single byte of data leaving the device.
Demo Video -> Click Here
Downloadable APK -> Click Here
Application Interface
- Zero-latency inference — no round-trip to cloud servers
- Privacy-first — images and audio never leave the device(100% privacy)
- Works in remote areas — no cell data or WiFi needed
- INT8 quantized models — fast on low-end Android devices (1 GB RAM compatible)
- Capture or select an image from the camera roll
- On-device image classification using AnyWhere SDK + TFLite
- Returns: disease name, confidence score, symptoms, causes, step-by-step treatment, prevention tips
- Healthy crop detection — correctly identifies healthy crops and returns a "No Disease Detected" result (no false positives)
- Record up to 10 seconds of machinery audio
- On-device MFCC extraction + audio classification
- Detects: engine knock, bearing noise, exhaust issues, overheating, hydraulic failure, thresher blockage
- Auto-stops and diagnoses after 10 seconds
- 20+ offline entries covering:
- Crop diseases (Leaf Blast, Powdery Mildew, Early Blight, Aphids, Stem Borer, Downy Mildew, Whitefly)
- Soil problems (Salinity, Compaction, Iron Deficiency, Acidity, Phosphorus Deficiency)
- Machinery issues (Engine Knock, Overheating, Hydraulic Failure, Thresher Blockage)
- Irrigation problems (Drip Clogging, Waterlogging, Sprinkler Malfunction, Canal Seepage)
- Full-text search, category filtering, severity badges
- Last 50 diagnoses stored locally with AsyncStorage
- Tap any entry to revisit full report
- Delete individual entries or clear all
| Language | Code | Native Name |
|---|---|---|
| English | en |
English |
| Hindi | hi |
हिंदी |
| Marathi | mr |
मराठी |
| Tamil | ta |
தமிழ் |
| Telugu | te |
తెలుగు |
| Punjabi | pa |
ਪੰਜਾਬੀ |
- Light / Dark / System (follows device)
- Instant toggle — works on both Android and iOS
- Persisted across app restarts
| Layer | Technology |
|---|---|
| Framework | Expo SDK 54 + React Native |
| Routing | Expo Router (file-based) |
| AI Engine | AnyWhere SDK |
| On-device ML | TensorFlow Lite (Android), CoreML (iOS) |
| State | React Context + AsyncStorage |
| Animations | React Native Reanimated 3 |
| Fonts | Nunito (Google Fonts) |
| Icons | @expo/vector-icons (Ionicons) |
| Audio | expo-av (MFCC pipeline) |
| Camera | expo-image-picker |
Farmphile AI uses the AnyWhere SDK by RunanywhereAI to perform all neural network inference directly on the device. Use of RunAnywhere SDK
Farmphile AI integrates the RunAnywhere SDK to enable efficient on-device AI inference directly on Android devices. The SDK allows machine learning models to run locally on the smartphone without requiring any cloud connectivity.
-
Crop and Soil Image Analysis Images captured using the device camera are processed locally through an on-device model to detect crop diseases, pest damage, and nutrient deficiencies.
-
Agricultural Machinery Sound Diagnosis Audio recordings of farming machinery are analyzed using an on-device audio classification model to detect abnormal sounds, potential mechanical failures, and severity levels.
The SDK helps optimize model execution for low latency and efficient resource usage, making it possible to deliver AI-powered analysis even on low-end Android devices commonly used in rural areas.
By leveraging RunAnywhere’s local inference capabilities, Farmphile AI ensures:
100% offline functionality
Fast real-time predictions
No cloud infrastructure dependency
Complete user data privacy
This integration demonstrates how AI can be deployed directly on edge devices to create scalable and accessible solutions for real-world problems such as agricultural diagnostics in connectivity-limited regions.
// Crop Disease — image classification
// constants/knowledgeBase.ts → diagnoseCropImage()
const sdk = new AnyWhereSDK({ modelPath: 'crop_disease_v2.tflite' });
const result = await sdk.classifyImage(imageUri);
const knowledge = mapResultToKnowledge(result.label);
// Sound Diagnosis — audio classification
// constants/knowledgeBase.ts → diagnoseMachinerySound()
const mfcc = extractMFCC(audioBuffer, { sampleRate: 16000, numCoeffs: 40 });
const sdk = new AnyWhereSDK({ modelPath: 'machinery_sound_v1.tflite' });
const result = await sdk.classify(mfcc);farmphile-ai/
├── app/
│ ├── _layout.tsx # Root layout with providers
│ ├── (tabs)/
│ │ ├── _layout.tsx # Tab bar layout
│ │ ├── index.tsx # Home (history + stats)
│ │ ├── crop.tsx # Crop analysis screen
│ │ ├── sound.tsx # Sound diagnosis screen
│ │ └── knowledge.tsx # Knowledge base browser
│ ├── settings.tsx # Settings (theme, language, privacy)
│ └── result/[id].tsx # Diagnosis result detail
├── context/
│ └── AppContext.tsx # Theme, language, diagnosis history
├── constants/
│ ├── knowledgeBase.ts # 20+ offline entries + AI inference hooks
│ ├── translations.ts # 6-language translations
│ └── colors.ts # Light/dark color palette
├── components/
│ └── ErrorBoundary.tsx # Crash recovery
├── android/
│ ├── build.gradle # Root Gradle — AnyWhere SDK + TFLite
│ ├── app/
│ │ ├── build.gradle # App Gradle — dependencies, proguard
│ │ ├── src/main/
│ │ │ ├── AndroidManifest.xml # NO internet permission
│ │ │ └── res/xml/
│ │ │ └── network_security_config.xml # Blocks all traffic
│ │ └── proguard-rules.pro
│ └── gradle/
│ └── wrapper/
│ └── gradle-wrapper.properties
└── README.md
The project includes complete Android Studio-compatible Gradle files. Models are loaded from android/app/src/main/assets/models/.
// android/app/build.gradle — key dependencies
implementation 'org.tensorflow:tensorflow-lite:2.13.0'
implementation 'org.tensorflow:tensorflow-lite-support:0.4.4'
// AnyWhere SDK: github.com/RunanywhereAI/runanywhere-sdks
implementation 'com.runanywhere:sdk-android:1.0.0'<!-- AndroidManifest.xml — NO internet permission granted -->
<!-- <uses-permission android:name="android.permission.INTERNET" /> -->
<!-- network_security_config.xml — blocks all network traffic -->
<network-security-config>
<base-config cleartextTrafficPermitted="false">
<trust-anchors />
</base-config>
</network-security-config>Copy your TFLite models to:
android/app/src/main/assets/models/
├── crop_disease_v2.tflite # Crop disease classifier (INT8)
└── machinery_sound_v1.tflite # Audio event classifier (INT8)
- Node.js 18+
- Expo CLI:
npm install -g @expo/cli - Expo Go app on your Android/iOS device
# Clone the repository
git clone https://github.com/your-username/farmphile-ai.git
cd farmphile-ai
# Install dependencies
npm install
# Start the development server
npm run expo:dev
# Scan the QR code with Expo Go# Build APK via Expo EAS
eas build --platform android --profile production
# Or open in Android Studio
# File → Open → android/Rice / Paddy · Wheat · Sugarcane · Tomato · Potato · Cotton · Grapes · Cucumber · Millet · Maize · Soybean · Chilli · Peas · All vegetables
| Feature | Status |
|---|---|
| Internet Permission | ❌ Disabled |
| Cloud Storage | ❌ None |
| Analytics SDK | ❌ None |
| Data Sharing | ❌ None |
| On-device Inference | ✅ AnyWhere SDK |
| Local Storage Only | ✅ AsyncStorage |
| Images Stay on Device | ✅ Always |
Built for: THE GPT CHALLENGE by Guru Gobind Singh Indraprastha University (GGSIPU), Delhi
Track: Agricultural Technology / AI for Good
Team: CODEX
Prakruti Parimita Rath(Team Lead)
Suraj Nath
Manish Debnath
Deep Ball
MIT License — see LICENSE for details.
- AnyWhere SDK by RunanywhereAI — on-device ML inference
- Expo — React Native framework
- TensorFlow Lite — Mobile ML runtime
- ICAR — Agricultural disease reference data