Azure MLOps

This project showcases a comprehensive MLOps pipeline using Azure services. Here's a breakdown of how the project was implemented:

Architecture and Deployment

This is the app deployed on Azure Cloud under the following architecture:

App

The app has 2 components. Frontend and Backend. In the image below you can see how the app works:

Model Training and Tracking

• Trained a heart disease prediction model using Azure Machine Learning.
• Experiment tracking and model versioning were managed via MLflow, facilitating reproducibility and model comparison.

API Development with FastAPI

• Implemented a FastAPI-based API to perform real-time inferences.
• Utilized the MLflow registered model hosted on the Azure tracking server, ensuring easy model deployment and inference.

Streamlit Frontend

• Developed a Streamlit-based user interface, enabling interaction with the API for real-time inference.

Containerization with Docker

• Created Docker images for each service.
• Uploaded these images to Azure Container Registry, ensuring a centralized repository for container images.

Deployment with Azure Web App

• Deployed an Azure Web App configured to pull the Docker images from Azure Container Registry and host the application.

Continuous Integration/Continuous Deployment (CI/CD)

• Established an end-to-end CI/CD pipeline to automate the workflow from running tests to deploying the application upon new changes.

Scheduled Batch Inference

• Orchestrated a scheduled workflow that fetches and preprocesses data on a monthly basis.
• This workflow then pulls the registered MLflow model from Azure, performs inferences on the records, and uploads the results to Azure Blob Storage.

This project serves as a comprehensive example of MLOps using Azure, covering model training, deployment, CI/CD automation, and scheduled data processing, offering a scalable and efficient approach for machine learning lifecycle management.