AI-Powered Agricultural Operating System for the Indian Farming Ecosystem
AgriSage is a unified, intelligent farming platform that integrates computer vision, real-time market intelligence, voice-enabled AI, and deep learning to empower farmers with actionable insights, yield optimization, and sustainable practices.
graph TD
User([Farmer/User]) --> Auth{Auth / Guest}
Auth --> Dashboard{Dashboard}
Dashboard --> Scanner[Crop Scanner]
Dashboard --> Advisor[Voice AI Advisor]
Dashboard --> Market[Market Intel]
Dashboard --> Sustainable[Sustainable Portal]
Dashboard --> ML[GenAffNet Hub]
Dashboard --> Learning[Learning Portal]
Dashboard --> SQLite[(SQLite DB)]
Scanner --> GeminiV[Gemini Vision AI]
Advisor --> WebSpeech[Web Speech API]
Advisor --> GeminiC[Gemini NLP]
Market --> WeatherAPI[WeatherAPI.com]
Market --> MandiAPI[Mandi Price Engine]
Sustainable --> CarbonTrack[Carbon Credit Tracker]
Sustainable --> WasteMap[Waste Collection Map]
WasteMap --> GoogleMaps[Google Maps API]
ML --> DeepLearning[GenAffNet Model]
GeminiV --> Results[Diagnosis & Recs]
DeepLearning --> Yield[Yield Prediction]
CarbonTrack --> Credits[Carbon Credits]
subgraph Services
ExpressBackend[Express Server]
CacheService[(Local Cache)]
end
ExpressBackend --> SQLite
Results -.-> CacheService
A comprehensive bar-graph analytics dashboard providing a "Regional Pulse" of India's agricultural performance. Features integrated satellite telemetry and mandi reports for North, West, Central, East, and South India.
Real-time Google Maps integration to locate verified stubble recycling, composting, and biomass energy facilities. Helps farmers monetize agricultural waste and reduce environmental impact.
Identify pests and diseases instantly using your device's camera. Leveraging 3D-CNN streams for spatial-spectral fusion.
Dataset Integration: The scanner leverages a comprehensive agricultural dataset featuring multiple crop images captured from various different angles to ensure robust classification. 🔗 View the Training Dataset
Note: For extra support and significantly more accurate diagnostic results, you can integrate your Gemini API key to unlock advanced cloud-based analysis.
Live Mandi prices for major commodities (Wheat, Paddy, Cotton, etc.) scraped and structured in real-time from trusted national sources.
5-day hyper-local forecasts with specific guidance on irrigation and harvesting windows based on humidity and wind trends.
A multilingual, voice-enabled assistant that provides science-backed agricultural advice in regional dialects.
- Waste Exchange: Connect with biomass energy plants to monetize farm stubble.
- Carbon Credits: A conceptual ledger for earning credits through sustainable practices.
A curated, multilingual video library designed to enhance agricultural expertise. Features 25+ hand-picked tutorials on crop cultivation (Wheat, Rice, Corn, Cotton), market price analysis, sustainable stubble management, and disease prevention techniques.
- Frontend: React 18+ / TypeScript / Vite
- Styling: Tailwind CSS (Dark Mode, Responsive, Glassmorphism)
- Animations: Framer Motion
- Backend: Node.js / Express / JWT Authentication
- Database: SQLite (Persistent storage via
better-sqlite3) - AI/ML: Google Gemini 2.0 Flash + GenAffNet Deep Learning Model
- Maps: Google Maps JS API (@react-google-maps/api)
- APIs: Google Generative AI SDK, WeatherAPI.com, Web Speech API
- Icons: Lucide React
- Storage: Persisted SQLite for Users/Activities & LocalStorage with 1-hour TTL Caching
- Node.js 18+
- npm or yarn
# Clone the repository
git clone https://github.com/TanayKapoor21/AGRISAGE.git
cd AGRISAGE
# Install dependencies
npm install
cd server && npm install && cd ..
# Copy environment variables
cp .env.example .env
# Add your API keys in .env
# VITE_GEMINI_API_KEY=your_gemini_key
# VITE_WEATHER_API_KEY=your_weather_key
# VITE_GOOGLE_MAPS_API_KEY=your_google_maps_key
# Start frontend (Tab 1)
npm run dev
# Start backend (Tab 2)
npm run serverThe frontend will open at http://localhost:5173 and the backend will run at http://localhost:5000
The app works fully without API keys using intelligent mock data. For live AI/Map features:
| API | Get Key At | Used For |
|---|---|---|
| Google Gemini | Google AI Studio | Vision, Chat, Predictions |
| WeatherAPI | weatherapi.com | Weather Forecasting |
| Google Maps | GCP Console | Interactive Waste Facility Map |
agrisage/
├── src/ # Frontend source code
│ ├── components/ # Layout, Sidebar, Header, WasteCollectionMap
│ ├── context/ # AppContext (Auth state, Theme, Language)
│ ├── pages/ # Dashboard, Auth, Scanner, Advisor, Market, Sustainable, Learning Portal, etc.
│ ├── services/ # Gemini AI, Weather API, Local/Backend Sync
│ └── types/ # TypeScript interfaces
├── server/ # Backend Node/Express server
│ ├── index.js # Express routes (Auth, Activities)
│ ├── db.js # SQLite Schema & Connection
│ └── agrisage.db # Persistent SQL Database
├── public/ # Static assets (logo, etc.)
├── package.json # Dependencies & Scripts
└── vite.config.ts # Frontend build config
- Colors: Sage (green), Harvest (gold), Earth (neutral)
- Effects: Glassmorphism, gradient cards, micro-animations
- Accessibility: High-contrast mode, icon-only navigation
- Fonts: Inter (body), Outfit (headings)
- Theme: Dark mode default with light mode toggle
When API quotas are reached, AgriSage doesn't break:
- Shows "High-Accuracy Regional Estimates" with realistic mock data
- Displays an API Status Banner to inform the user
- Map shows a decorative fallback with static markers and instructions
- All data is cached in localStorage with 1-hour TTL
- Phase 2: IoT Soil Sensor integration for automated real-time alerts.
- Phase 3: Blockchain-linked Carbon Credit verification system.
- Phase 4: Expansion to 12+ regional languages with localized dialect support.
AgriSage implements the GenAffNet, a state-of-the-art inference engine based on our research paper for plant disease detection. It utilizes Hyperspectral Imaging (HSI) and Generative AI to provide precision diagnostics.
The diagnostic pipeline is structured into four distinct phases as validated in our sugarbeet dataset:
- Hyperspectral Preprocessing: Automated PCA-based band reduction (from 224 to 96 bands) and 9×9 spatial patch extraction to eliminate redundancy.
- DMLPFFN Architecture: A hierarchical spectral-spatial fusion network using Global, Partition, and Local Perceptrons with dilated convolutions.
- GenAI Augmentation: Integration of a Convolutional VAE (Variational Autoencoder) to synthesize high-diversity spectral patterns, achieving 98.09% precision.
- Diagnostic Mapping: Automated classification into four classes: Healthy/Early Stress, Fungal, Bacterial, and Viral-Nematode manifestations.
graph TD
Cube[HSI Cube .NPY] --> PCA[PCA Band Reduction]
PCA --> Patch[9x9x96 Patch Extraction]
subgraph "DMLPFFN Block"
Patch --> GP[Global Perceptron]
Patch --> PP[Partition Perceptron]
Patch --> LP[Local Perceptron]
GP --> Fusion[Feature Fusion]
PP --> Fusion
LP --> Fusion
end
subgraph "VAE Augmentation"
CVAE[Conv-VAE] --> Synth[Synthetic Patterns]
Synth -.-> Fusion
end
Fusion --> Voting[Patch-Level Voting]
Voting --> Classify{Diagnostic Result}
Classify --> H[Healthy]
Classify --> F[Fungal]
Classify --> B[Bacterial]
Classify --> V[Viral/Nematode]
🔗 Full Research Code & Results: sugarbeet-genai Repository
- Frontend: React 18+ / TypeScript / Vite
- Styling: Tailwind CSS (Dark Mode, Responsive, Glassmorphism)
- Animations: Framer Motion
- Backend: Node.js / Express / JWT Authentication
- Database: SQLite (Persistent storage via
better-sqlite3) - AI/ML: Google Gemini 2.0 Flash + GenAffNet Deep Learning Model
- Maps: Google Maps JS API (@react-google-maps/api)
- APIs: Google Generative AI SDK, WeatherAPI.com, Web Speech API
- Icons: Lucide React
- Storage: Persisted SQLite for Users/Activities & LocalStorage with 1-hour TTL Caching
AgriSage serves as the primary implementation platform for our research on Precision Sugarbeet Cultivation. Using an advanced iteration of the GenAffNet (Agricultural Affinity Network) deep learning model, we optimize nitrogen application and identify diseases with elite accuracy.
| Name | Primary Focus | Research Contribution |
|---|---|---|
| Tanay Kapoor | Core AI Architecture & Integration | Lead Model Training & DMLPFFN Optimization |
| Akash Yadav | System Logic & Data Pipeline | Hyperspectral Preprocessing & PCA Band Reduction |
| Kanika Yadav | UX Strategy & Frontend Design | UI Diagnostic Mapping & Stage-based Workflow |
| Srasthti Chauhan | Agricultural Intelligence & Data Analysis | Dataset Validation & Precision Metric Analysis |
Special thanks to Dr. Anuradha Dhull and Dr. Asha Sohal for their scientific guidance and agricultural insights.
MIT License — Built with ❤️ for Indian Agriculture