AgriMitra is a complete, end-to-end AI-powered smart agriculture platform designed to support Indian farmers and agricultural authorities with data-driven, localized, and accessible decision-making.
Built for Smart India Hackathon (SIH) 2025 – Problem Statement ID 25044, AgriMitra integrates Machine Learning, Agentic AI, IoT sensing, and Satellite Intelligence to deliver yield prediction, optimization, disease detection, and governance-level monitoring at scale.
Philosophy: Beyond a tool, a trusted digital companion that understands every farmer, their land, their crops, and their language, irrespective of their level of education or digital literacy.
-
- Crop Yield Prediction using real-time soil, weather, and historical data
-
- Agentic AI Optimization for fertilizer, irrigation, and soil health
-
- Satellite-Based Farm Intelligence (NDVI, crop stress, vegetation indices)
-
- Disease & Pest Detection using deep learning
-
- Voice-Enabled Multilingual Farmer App
-
- Fertilizer Authenticity Verification via barcode validation
-
- Government Monitoring Dashboard for data-driven policy and advisories
AgriMitra follows a layered, modular architecture optimized for scalability and rural deployment constraints.
IoT Sensors / Satellite Data / Weather APIs / Farmer Inputs
↓
Data Ingestion Layer
↓
ML & Agentic Intelligence Layer
(Prediction | Optimization | Vision)
↓
Application Service Layer
(APIs | Business Logic)
↓
Farmer Mobile App & Govt Web Dashboard
Objective: Predict expected yield before harvest to guide early interventions.
Inputs
- Soil parameters: NPK, moisture, pH
- Weather data: rainfall, temperature, humidity
- Crop metadata: type, sowing date, region
- Historical yield records
Main Model Used
- Random Forest Regressor
Why Random Forest?
- Handles non-linear agricultural relationships
- Robust to noisy, incomplete field data
- Interpretable feature importance
Objective: Convert predictions into actionable farming decisions.
Capabilities
- Fertilizer dosage optimization
- Soil correction recommendations
- Irrigation scheduling
- Crop-specific advisory
Design
- Rule-guided AI agents augmented with ML predictions
- Continuous feedback loop from updated sensor and weather data
Objective: Enable farm health monitoring without mandatory IoT hardware.
Technology
- Google Earth Engine (GEE)
Outputs
- NDVI and vegetation health maps
- Nutrient stress indicators
- Crop growth trend analysis
Impact
- Scales to sensor-scarce rural regions
- Reduces hardware dependency
Objective: Early identification of crop diseases and pests from images.
Approach
- Computer Vision using Deep Learning
Models
- Convolutional Neural Networks (CNNs)
- EfficientNet for accuracy–efficiency tradeoff
Datasets
- PlantVillage
- PlantDoc
- Kaggle agricultural datasets
Outputs
- Disease/pest classification
- Symptoms
- Recommended treatment and pesticides
- Helpline and advisory guidance
Objective: Maximize adoption across literacy and language barriers.
Features
- Voice input and output
- Regional language support (English, Hindi, Tamil, Odia)
- Minimal, icon-driven UI
Result
- High accessibility for small and marginal farmers
Problem Addressed: Counterfeit fertilizers impacting yield and income.
Solution
- Barcode scanning
- Validation against government-approved databases
Stakeholders
- Agricultural departments
- Policy makers
Capabilities
- Yield performance analytics
- Pest outbreak heatmaps
- Crop distribution insights
- Subsidy and scheme monitoring
- Data-driven advisory generation
- Cloud-hosted backend architecture
- API-first, service-oriented design
- Offline and low-connectivity support
- Scalable from village to national level
| Challenge | Mitigation Strategy used in AgriMitra |
|---|---|
| Data inconsistency | Hybrid global + local datasets |
| Low digital literacy | Voice-first UX + simple UI |
| Connectivity gaps | Offline-capable workflows |
| Sensor affordability | Govt/NGO subsidy models |
-
- Economic: Improved yield and farmer income
-
- Social: Inclusive access regardless of literacy
-
- Environmental: Optimized fertilizer and water usage
-
- Governance: Data-driven agricultural planning
-
- ML / DL: Scikit-learn, TensorFlow / PyTorch
-
- Satellite: Google Earth Engine
-
- Backend: Python, REST APIs
-
- Frontend: Mobile app + Web dashboard
-
- Data: Open agricultural datasets + real-time APIs
- YouTube Demo: https://youtu.be/vdfzczCk864
- GitHub Repository: https://github.com/SIH-Project/Agri-Mitra
Developed by a multidisciplinary student team, including Atharva Shukla, Sagar Awasthi, Michelle Ellen Joseph, Uday Trivedi, Eshaan Adyanthaya, and Divyanshu Karmakar as part of Smart India Hackathon 2025, achieving a place among the Top 20 software teams selected at the college level.