This project demonstrates a MLOps pipeline for deploying a machine learning model into a production-ready web application. The goal is to help an insurance company forecast patient charges using input like age, gender, BMI, number of children, and smoking status.
The solution includes:
- A machine learning model trained to predict insurance charges
- A Flask back-end to serve predictions
- A HTML / CSS front-end for user input
- A CI/CD pipeline with Github Actions
- Unit tests with Pytest
- Containerization with Docker
- Automatic cloud deployment on Microsoft Azure
- A product documentation
Architecture diagram
- Build the image
docker build -t insurancemodel.azurecr.io/mlops-insurance-prediction:latest .
- Run the container
docker run -d -p 5000:5000 insurancemodel.azurecr.io/mlops-insurance-prediction
The web app is served on http://localhost:5000/.
- Clone the repository
git clone https://github.com/PierreExeter/MLOps-Pipeline-Deployment
- Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh
- Create and activate the environment
uv venv --python 3.11
source .venv/bin/activate
- Install the dependencies
uv pip install -r requirements.txt
- Train the model
python src/train_model.py
- Run the app locally
# development
python src/app.py
# production
python src/app.py --production
The web app is served on http://localhost:5000/.