🚗 Vehicle Insurance ML Project — End-to-End MLOps on Azure

Hey there! 👋 Welcome to my Vehicle Insurance Machine Learning project — where I built everything from scratch using core MLOps principles without relying on any automation tools like DVC or Prefect.

This project is my personal deep dive into how real-world machine learning systems are built, deployed, and maintained in production. I wanted to understand the "hard parts" of MLOps by actually doing them myself — no shortcuts.

---

🧠 What’s This Project About?

The goal is simple: predict whether a customer is likely to take vehicle insurance, based on their profile.

But the real focus of this project is not just the prediction. It's about building a production-ready machine learning pipeline that covers:

• Clean data flow from MongoDB Atlas
• Full model lifecycle (train, validate, evaluate)
• CI/CD pipeline with GitHub Actions
• Deployment on Azure VM
• Model serving through a web app built with FastAPI
• Secure cloud storage via Azure Blob Storage

And most importantly — I wrote everything myself, following MLOps best practices as if I were part of a real team shipping a model to production.

---

🔧 Key Features

• Hardcoded, robust pipeline (no DVC or MLflow)

• End-to-end stages:

  • ✅ Data Ingestion (from MongoDB)
  • ✅ Data Validation (with schema checks)
  • ✅ Data Transformation (scaling, encoding, cleaning)
  • ✅ Model Training & Evaluation
  • ✅ Model Pusher (uploads model.pkl to Azure Blob Storage)

• Web App with FastAPI

  • Pulls the model from Azure
  • Accepts input and displays predictions on the frontend

• CI/CD pipeline using GitHub Actions

  • Automatically builds, pushes Docker image to Azure Container Registry (ACR)
  • Deploys to Azure VM via SSH

• Secrets Management

  • Used GitHub secrets for all sensitive credentials

---

🌐 Technologies I Used

• Python (core language)
• scikit-learn, pandas, numpy (for ML)
• MongoDB Atlas (cloud database)
• Azure Blob Storage (to store model)
• FastAPI (backend for prediction service)
• Docker (for containerization)
• Azure VM + ACR (cloud hosting and container registry)
• GitHub Actions (CI/CD)

---

🛠 How It All Works

Here’s a quick overview of the pipeline:

graph TD;
    A[MongoDB Atlas] --> B[Data Ingestion]
    B --> C[Data Validation]
    C --> D[Data Transformation]
    D --> E[Model Training]
    E --> F[Model Evaluation]
    F --> G[Model Pusher to Azure Blob]
    G --> H[FastAPI Web App]
    H --> I[User Prediction]

Loading

Each step is a Python module that I wrote to follow software engineering practices like modularization, exception handling, and logging.

---

🚀 How to Run This Project (End-to-End)

🧩 Note: This is a real-world deployment, so you’ll need some Azure setup.

🔐 GitHub Secrets Required

Make sure to set the following secrets in your GitHub repository:

• ACR_LOGIN_SERVER
• ACR_USERNAME
• ACR_PASSWORD
• MONGODB_URL
• AZURE_STORAGE_CONNECTION_STRING
• VM_IP

🔧 Azure VM Setup

• Open ports:

  • 22 (SSH)
  • 80 (HTTP)
  • 5000 (FastAPI App)

• Add a user and assign them Owner or Contributor role

📦 Run Locally (Optional)

python app.py

This will trigger the fastAPI app.

🌍 Visit the App

Once deployed, go to:

http://<your-azure-vm-ip>:5000/

You’ll see the FastAPI Web app

---

💡 Why I Built It This Way

I could’ve used DVC, MLflow, or other tools — but I didn’t. Why?

Because I wanted to really understand what’s going on under the hood. I’ve seen how ML engineers rely on tools that abstract away the pipeline, and I wanted to do the opposite.

This project helped me:

• Think like a software engineer and not just a data scientist
• Learn how CI/CD works with real cloud infrastructure
• Get hands-on with secure, modular MLOps design

It was tough at times, but it made me a stronger engineer overall.

---

🧑‍💼 About Me

I’m someone who loves figuring things out from the ground up.

• 🔍 I enjoy solving real problems — not just getting models to work.
• 🛠 I care about clean code, clear structure, and maintainability.
• ☁️ I’m passionate about cloud technologies and real-world deployment.
• 🤝 I value teamwork, ownership, and continuous learning.

I’m currently looking for remote opportunities (internship, freelance, or full-time) in ML Engineering, MLOps, or Backend Dev.

---

📬 Let’s Connect!

• 📧 Email: farjhanahmed@gmail.com
• 🔗 LinkedIn: in/farjhan-ahmed
• 💻 Portfolio: portfolio

---

🙏 Thanks for checking out my project!

This wasn't just about building an ML model — it was about building my confidence and proving to myself that I can take an idea from data to deployment, all by myself.

If you’re hiring, mentoring, or just want to geek out about MLOps or FastAPI, I’d love to connect. 😊

---