Krishi Saarthi

Agentic AI-powered agricultural advisor platform serving India's 120 million farmers. Provides hyperlocal farming guidance, disease diagnosis, market intelligence, and crisis support through specialized AI models.

Long Form Synopsis:

The Misfits - Paper: https://dub.sh/TheMisfitsSynopsisMain

Demo Video:

Krishi Saarthi - Demo Video: https://dub.sh/TheMisfitsDemo

Architecture Diagram of the Main Agent:

Repository Structure

Folder	Description
backend/	Python FastAPI backend with AI/ML orchestration
mobile/	Flutter cross-platform mobile application
core-ml/	ML models submodule hosted on Lightning AI CUDA server
mlscripts/	Training scripts for QLoRA fine-tuning, instruction-tuned TinyLlama, LSTM models

Architecture

• Backend: Python FastAPI with AI/ML orchestration
• Mobile: Flutter cross-platform application
• AI Models: LoRA-tuned VLM, Kisan Call Centre prompt-tuned TinyLlama, LSTM forecasting, Kisan Call Centre prompt-tuned TinyLlama, RAG based Agricultural scheme advisor
• Data Sources: Government APIs, KCC transcripts, agricultural datasets

Prerequisites

Backend Requirements

• Python 3.12
• Poetry (dependency management)

Mobile Requirements

• Flutter SDK 3.0+
• Dart SDK
• Android Studio / VS Code
• Android SDK (for Android development)
• Xcode (for iOS development on macOS)

Backend Setup

1. Install Poetry

# macOS/Linux
curl -sSL https://install.python-poetry.org | python3 -

# Windows (PowerShell)
(Invoke-WebRequest -Uri https://install.python-poetry.org -UseBasicParsing).Content | python -

2. Clone Submodules

git submodule update --init --recursive

3. Setup Backend

cd backend
poetry config virtualenvs.in-project true
poetry install --no-root

4. Run Backend

# Option 1: With Poetry
poetry run python main.py

# Option 2: Activate environment first
poetry shell
python main.py

# Option 3: With uvicorn
poetry run uvicorn main:app --reload

Backend will be available at http://localhost:8000 API documentation at http://localhost:8000/docs

Mobile Setup

1. Install Flutter

Windows

# Download Flutter SDK from https://flutter.dev/docs/get-started/install/windows
# Extract and add to PATH

macOS

# Using Homebrew
brew install --cask flutter

# Or download from https://flutter.dev/docs/get-started/install/macos

Linux

# Download Flutter SDK
wget https://storage.googleapis.com/flutter_infra_release/releases/stable/linux/flutter_linux_3.16.0-stable.tar.xz
tar xf flutter_linux_3.16.0-stable.tar.xz
export PATH="$PATH:`pwd`/flutter/bin"

2. Verify Flutter Installation

flutter doctor

3. Setup Mobile App

cd mobile
flutter pub get

4. Run Mobile App

# List available devices
flutter devices

# Run on specific device
flutter run -d <device_id>

# Run on Android emulator
flutter run

# Run on iOS simulator (macOS only)
flutter run -d ios

Development Workflow

Start Both Applications

# Terminal 1 - Backend
cd backend
poetry shell
python main.py

# Terminal 2 - Mobile
cd mobile
flutter run

API Testing

Import the Postman collection for comprehensive API testing:

• Collection: backend/krishi_saarthi_production_postman_collection.json
• Base URL: http://localhost:8000
• Endpoints: Chat processing, Twilio integration, mandi prices, government schemes, KCC advice, YouTube recommendations, guided farming mode

Quick API Endpoints

• Health Check: GET /health
• API Health: GET /api/health/
• Sample: GET /api/sample/
• Hello: GET /api/sample/hello/{name}

Project Structure

Folder	Description
backend/	Python FastAPI backend with AI/ML orchestration
mobile/	Flutter cross-platform mobile application
core-ml/	ML models submodule hosted on Lightning AI CUDA server
mlscripts/	Training scripts for QLoRA fine-tuning, instruction-tuned TinyLlama, LSTM models

krishi-saarthi/
├── backend/          # Python FastAPI backend
│   ├── main.py      # Application entry point
│   ├── pyproject.toml
│   └── app/         # Main application package
│       ├── api/     # API routes and endpoints
│       ├── core/    # Core configuration
│       ├── models/  # ML model loading and inference
│       ├── services/# Business logic services
│       ├── utils/   # Utility functions
│       └── schemas/ # API schemas
├── mobile/          # Flutter mobile application
│   ├── pubspec.yaml
│   └── lib/
├── core-ml/         # ML models submodule (Lightning AI CUDA server)
│   ├── models/      # Pre-trained AI models
│   ├── training/    # Training scripts and datasets
│   └── inference/   # Model inference services
└── mlscripts/       # Training scripts for QLoRA fine-tuning, instruction-tuned TinyLlama, LSTM models

Troubleshooting

Backend Issues

• Ensure Python 3.12 is installed
• Run poetry install --no-root if dependencies fail
• Check poetry show for installed packages

Mobile Issues

• Run flutter doctor to check setup
• Ensure Android SDK is properly configured
• For iOS: Xcode must be installed on macOS
• Clear Flutter cache: flutter clean && flutter pub get

Key Features

• Vision-based crop disease diagnosis
• LSTM price forecasting and market intelligence
• Multilingual conversational AI agent
• SMS fallback via Twilio integration
• Crisis detection and safety protocols
• Government scheme matching