AgriAI

Cloud-Native Precision Agriculture AI Platform

Architecture Overview

AgriAI is a multi-agent microservices architecture deployed on AWS. It provides a unified farm analysis report by orchestrating three core ML services:

1. Disease Detection Service: CNN (EfficientNet-B3) for detecting 38 crop diseases (PlantVillage taxonomy). Includes GradCAM visual explainability.
2. Weather Prediction Service: TCN (Temporal Convolutional Network) for 24-hour forecasting using historical Open-Meteo data. Uses MC Dropout for uncertainty estimation.
3. Market Analytics Service: XGBoost and Ridge Regression models predicting crop yields and market prices, combined with an SLSQP portfolio optimizer for land allocation.
4. Orchestration Agent: FastAPI service managing parallel, fault-tolerant execution of the above services. Synthesizes data into an actionable FarmInsightReport.

Directory Structure

• config/: Shared configuration and constants.
• shared/: Common utilities (JWT Auth, Logging, AWS Clients).
• services/: Microservices (Disease, Weather, Market, Orchestrator).
• infrastructure/: Terraform configurations and Lambda processors.
• pipelines/: Data acquisition and retraining scripts.
• tests/: Pytest suite.
• scripts/: Deployment and model download helpers.

Quickstart (Local Development)

1. Download Models:

   ./scripts/download_models.sh

   (Note: This creates mock models locally. In production, it downloads from S3).

2. Start LocalStack and Services:

   docker-compose up -d

3. Run Tests: Ensure AWS_ENDPOINT_URL=http://localhost:4566 is set.

   pytest tests/

API Documentation

• Orchestrator: http://localhost:8000/docs
• Disease Detection: http://localhost:8001/docs
• Weather Prediction: http://localhost:8002/docs
• Market Analytics: http://localhost:8003/docs

Example Orchestrator Request

curl -X POST "http://localhost:8000/analyze/farm" \
  -H "Authorization: Bearer <token>" \
  -F "image=@/path/to/leaf.jpg" \
  -F "sensors=@/path/to/sensors.csv" \
  -F 'farm_metadata={"crop_type": "corn", "soil_quality": 85.0, "fertilizer_kg_ha": 120.0, "water_mm": 400.0}'

Deployment to AWS

1. Configure AWS Credentials.
2. Run deployment script:

   ./scripts/deploy.sh

   This applies Terraform and pushes Docker images to ECR.