AI for the Fields
Agriculture employs over 40% of India's workforce, yet access to expert agricultural advice remains a challenge for millions of farmers. Enter Garuda, India's first Large Language Model designed specifically for agriculture—a $2 million project funded by Google.
The project aims to bring AI-powered advisory services to farmers in their native languages, covering everything from crop selection to market prices.
Project Overview
Core Objectives
| Objective | Description |
|---|---|
| Language Coverage | 12 major Indian languages |
| Knowledge Domains | Crops, weather, markets, schemes |
| Access Modality | Voice-first, SMS fallback |
| Target Users | Smallholder farmers |
| Deployment | Q3 2026 |
Funding and Partners
- Primary Funding: $2 million from Google
- Research Partner: ANNAM.AI at IIT Ropar
- Government Support: Ministry of Agriculture
- Implementation: State agriculture departments
Technical Architecture
Data Sources
Garuda is being trained on a diverse corpus:
Agricultural Knowledge
- ICAR (Indian Council of Agricultural Research) publications
- State agricultural university research
- Krishi Vigyan Kendra (KVK) advisories
- PM-KISAN scheme documentation
Real-Time Data
- IMD weather forecasts
- Mandi (market) price data
- Crop calendar information
- Pest and disease alerts
Traditional Knowledge
- Indigenous farming practices
- Regional crop varieties
- Folk weather prediction methods
Model Specifications
Base Model: Derived from open-source LLM
Fine-tuning: Agricultural domain adaptation
Languages: Hindi, Punjabi, Tamil, Telugu, Kannada,
Marathi, Bengali, Gujarati, Odia,
Malayalam, Assamese, Bhojpuri
Input: Voice (primary), Text (secondary)
Output: Voice response, SMS advisory
Response Time: < 3 seconds
Accuracy Target: 92% for factual queries
Use Cases
Crop Advisory
Farmer Query: "मेरे गेहूं की पत्तियां पीली हो रही हैं" (My wheat leaves are turning yellow)
Garuda Response: Analysis of symptoms, possible causes (nitrogen deficiency, waterlogging, rust disease), recommended actions, and nearby input dealer information.
Market Intelligence
Farmer Query: "कल मंडी में टमाटर का क्या भाव रहेगा?" (What will tomato prices be at the market tomorrow?)
Garuda Response: Price prediction based on arrival patterns, nearby mandi prices, best selling timing recommendation.
Scheme Information
Farmer Query: "PM-KISAN के लिए क्या करना होगा?" (What do I need for PM-KISAN?)
Garuda Response: Eligibility criteria, document requirements, application process, nearest Common Service Center.
Weather-Based Advice
Farmer Query: "इस हफ्ते खेत में क्या काम करें?" (What farm work should I do this week?)
Garuda Response: Weather forecast, recommended activities (sowing, irrigation, spraying), caution alerts.
Deployment Strategy
Phase 1: Pilot (Q1 2026)
- 3 states: Punjab, Andhra Pradesh, Maharashtra
- 10,000 farmers
- Voice hotline access
- Feedback collection
Phase 2: Expansion (Q2-Q3 2026)
- 10 states
- 100,000 farmers
- WhatsApp integration
- Farmer app launch
Phase 3: Scale (Q4 2026)
- National availability
- Integration with Kisan Call Centers
- Smart speaker deployment
- Agromet integration
Challenges Being Addressed
Linguistic Complexity
Indian agricultural terminology varies significantly by region:
| Standard Term | Regional Variations |
|---|---|
| Irrigation | सिंचाई, पाणी देणे, நீர்ப்பாசனம் |
| Fertilizer | खाद, उर्वरक, உரம் |
| Pest | कीट, किडा, பூச்சி |
Garuda is trained to understand and respond in local dialects.
Voice Interface Challenges
- Background noise in farm environments
- Code-mixing (Hindi-English)
- Unclear query formulation
- Network connectivity issues
Accuracy Requirements
Agricultural advice has direct financial impact:
- Wrong pest identification → crop loss
- Incorrect price information → financial loss
- Bad weather advice → spoiled harvest
Expected Impact
Projected Outcomes (Year 1)
| Metric | Target |
|---|---|
| Farmers Reached | 500,000 |
| Queries Answered | 10 million |
| Yield Improvement | 8-12% average |
| Cost Reduction | 15% on inputs |
| Income Increase | ₹8,000-15,000 per season |
Industry Significance
Garuda represents a new approach to agricultural AI:
- Vernacular-First: Built for Indian languages from scratch
- Domain-Specific: Depth over breadth in agriculture
- Voice-Primary: Designed for low-literacy users
- Offline-Capable: Works in low-connectivity areas
Looking Ahead
If successful, Garuda could become a template for sector-specific AI in developing countries. The project team is already documenting learnings for potential replication in:
- Fisheries
- Animal husbandry
- Forestry
- Rural healthcare
"Garuda isn't just an AI project—it's about bringing the power of knowledge to every farmer's fingertips, in a language and format they can use."
