🌾 KisanMind — Multi-Agent AI Farming Assistant for Indian Farmers

An intelligent multi-agent AI platform built with Google ADK that empowers Indian farmers with crop disease diagnosis, market intelligence, weather-aware farming recommendations, and government scheme discovery through collaborative AI agents.

---

🚜 Problem Statement

Over 140 million Indian farmers face fragmented access to agricultural knowledge.

They often struggle with

• Identifying crop diseases quickly
• Finding reliable mandi prices
• Planning around changing weather
• Discovering government support schemes

Most existing applications solve only one of these problems.

KisanMind brings them together inside one intelligent multi-agent ecosystem.

---

💡 Solution

KisanMind uses Google ADK to orchestrate specialized AI agents that collaborate to answer farmer queries using real-time APIs and multimodal reasoning.

Instead of relying on a single LLM, every request is routed to the most appropriate domain expert.

This results in

• higher accuracy
• lower latency
• modular scalability
• easier maintenance

---

🤖 Multi-Agent Architecture

                        User

                         │

                 Google ADK Orchestrator

      ┌──────────────┬──────────────┬──────────────┬──────────────┐
      │              │              │              │
 CropDoc       MarketMind      WeatherWatch    SchemeScout
      │              │              │              │
Gemini Vision   Agmarknet API   OpenMeteo API   Govt Schemes

                MCP Tool Server
         ├── Weather Tool
         ├── Market Tool
         └── Scheme Tool

---

✨ Features

🌿 Crop Disease Diagnosis

• Image upload
• Gemini Vision reasoning
• Automatic treatment recommendation
• Fertilizer suggestions

---

💰 Market Intelligence

• Live Agmarknet prices
• Nearby mandi comparison
• Price trends
• Better selling recommendations

---

🌦 Weather Intelligence

• 7-day forecast
• Crop-specific recommendations
• Rain alerts
• Irrigation planning

---

🏛 Government Schemes

Automatically finds

• PM-KISAN
• PMFBY
• KCC
• State subsidy programs

---

🧠 Agent Collaboration

Rather than answering directly,

the Orchestrator

✔ understands intent

✔ delegates work

✔ combines outputs

✔ produces one unified response

This follows Google's Agent Development Kit design philosophy.

---

🔌 MCP Tools

KisanMind demonstrates the Model Context Protocol (MCP) using reusable tools.

Tool	Purpose
Weather Tool	Weather forecasts
Market Tool	Agmarknet prices
Scheme Tool	Government schemes

Every agent accesses shared capabilities through MCP.

---

🔄 Automatic LLM Fallback

Primary Model

Gemini 3.5 Flash

Fallback

Groq
↓
Llama 3.3 70B

If Gemini experiences

• quota limits
• API failures
• temporary downtime

the request automatically retries on Groq without interrupting the user.

---

🔒 Security

• Environment variables
• No API keys in source code
• Input validation
• Secure image uploads
• File size limits
• MIME validation
• Pydantic request models

---

📦 Tech Stack

AI

• Google ADK
• Gemini 3.5 Flash
• LiteLLM
• Groq

Backend

• FastAPI
• Python

Frontend

• Streamlit

APIs

• Open-Meteo
• data.gov.in (Agmarknet)

Protocols

• MCP

---

🚀 Deployment

Supports deployment on

• Hugging Face Spaces
• Render
• Railway
• Docker
• Local Machine

---

📊 Future Roadmap

• Voice conversations in Hindi
• Offline edge inference
• WhatsApp farming assistant
• Pest outbreak prediction
• Satellite imagery analysis
• Personalized crop calendars
• IoT sensor integration

---

🎯 Kaggle AI Agents Course Concepts Demonstrated

Concept	Implementation
✅ Google ADK	Multi-Agent Orchestration
✅ MCP Server	Shared Tool Calling
✅ Deployability	FastAPI + Streamlit
✅ Security	Environment Variables + Validation
✅ Agent Skills	Specialized Domain Agents
✅ Multi-Agent Collaboration	Intent Routing

Astitva Bhardwaj AI Engineer • Machine Learning • Multi-Agent Systems