Reco Agricole

Crop yield prediction and recommendation system. Given field conditions (region, soil type, rainfall, temperature, pesticides, etc.), the API predicts the expected yield for a specific crop or recommends the highest-yielding option across Maize, Rice, and Wheat.

You can view the dashboard at https://webapp-982044430070.europe-west1.run.app/.

Stack

Layer	Technology
ML models	XGBoost
API	FastAPI + Uvicorn
Dashboard	Streamlit + Plotly
Container registry	Google Artifact Registry
Deployment	Google Cloud Run
CI/CD	GitHub Actions → Cloud Run

Project structure

.
├── src/
│   ├── models/             # Trained XGBoost model files
│   │   ├── maize_model.json
│   │   ├── rice_model.json
│   │   └── wheat_model.json
│   ├── models.py           # Model loading + validation + inference
│   ├── api.py              # FastAPI app
│   ├── webapp.py           # Streamlit dashboard
│   ├── tests.py            # pytest test suite
├── .streamlit/config.toml  # Streamlit config
├── notebooks/              # EDA, feature engineering, fine-tuning
├── data/
│   ├── raw/
│   └── processed/
├── Dockerfile.api
├── Dockerfile.webapp
├── docker-compose.yml
├── pyproject.toml
└── uv.lock

Local development

Requirements: Python 3.12, uv

# Install dependencies
uv sync

# Start the API
uv run fastapi run src/api.py

# In a second terminal, start the dashboard
uv run streamlit run src/webapp.py

The dashboard is available at http://localhost:8501. It expects the API at http://localhost:8000 by default.

Running tests

uv run pytest src/tests.py -v

Local Docker

# Build and start both services
docker compose up -d --build

# View logs
docker compose logs -f api
docker compose logs -f webapp

Service	URL
API	http://localhost:8000
API docs	http://localhost:8000/docs
Dashboard	http://localhost:8501

API endpoints

Method	Path	Description
GET	/health	Healthcheck
POST	/predict	Yield prediction for a specific crop
POST	/recommend	Best crop recommendation across all three

Predict payload example:

{
  "crop": "Rice",
  "data": {
    "rain (mm)": 534.0,
    "temp (C)": 14.74,
    "Year": 2013,
    "pesticides_tonnes": 45177.18,
    "Area": "Australia",
    "Days_to_Harvest": 104,
    "Irrigation_Used": true,
    "Fertilizer_Used": true,
    "Soil_Type": "Silt"
  }
}

Recommend payload example:

{
  "data": {
    "rain (mm)": 534.0,
    "temp (C)": 14.74,
    "Year": 2013,
    "pesticides_tonnes": 45177.18,
    "Area": "Australia",
    "Days_to_Harvest": 104,
    "Irrigation_Used": true,
    "Fertilizer_Used": true,
    "Soil_Type": "Silt"
  }
}

CI

GitHub Actions runs on every push and pull request to main:

1. Test — installs dependencies with uv sync --frozen, runs the full pytest suite
2. Smoke test — builds and starts the API via Docker Compose, waits for the healthcheck, then hits /health and /predict to verify the containerised build works end-to-end