🚗 Vehicle Insurance MLOps

An end-to-end MLOps project for predicting vehicle insurance cross-sell outcomes — from raw data ingestion to a Dockerized API deployed on AWS EC2 with a full CI/CD pipeline.

---

📌 Overview

This project implements a production-grade machine learning pipeline to predict whether a customer is likely to opt for vehicle insurance. It demonstrates best practices in MLOps: modular code structure, automated training pipelines, cloud model storage, and continuous deployment via GitHub Actions.

---

🏗️ Architecture

MongoDB Atlas (Raw Data)
        ↓
Data Ingestion → Data Validation → Data Transformation
        ↓
   Model Trainer → Model Evaluation → Model Pusher (AWS S3)
        ↓
   FastAPI Prediction App (Dockerized)
        ↓
   AWS EC2 via GitHub Actions CI/CD

---

⚙️ Tech Stack

Category	Tools
Language	Python 3.10
ML & Data	Scikit-learn, Pandas, NumPy
Database	MongoDB Atlas
Cloud Storage	AWS S3
Deployment	AWS EC2, Docker, ECR
CI/CD	GitHub Actions (Self-hosted Runner)
API	Flask / FastAPI
Environment	Conda

---

📁 Project Structure

vehicle-insurance-mlops/
├── src/
│   ├── components/          # Pipeline components (ingestion, validation, etc.)
│   ├── configuration/       # MongoDB & AWS connection configs
│   ├── cloud_storage/       # AWS S3 storage utilities
│   ├── data_access/         # Data fetch & transform layer
│   ├── entity/              # Config & artifact entities, estimator
│   ├── pipeline/            # Training pipeline orchestration
│   └── utils/               # Shared utilities
├── config/
│   └── schema.yaml          # Dataset schema for validation
├── notebook/                # EDA & feature engineering notebooks
├── static/ & templates/     # Frontend for prediction UI
├── app.py                   # Application entry point
├── demo.py                  # Pipeline test runner
├── Dockerfile
├── requirements.txt
├── setup.py
└── pyproject.toml

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/<your-username>/vehicle-insurance-mlops.git
cd vehicle-insurance-mlops

2. Set up the environment

conda create -n vehicle python=3.10 -y
conda activate vehicle
pip install -r requirements.txt

3. Configure environment variables

Bash:

export MONGODB_URL="mongodb+srv://<username>:<password>@cluster.mongodb.net/"
export AWS_ACCESS_KEY_ID="your-access-key"
export AWS_SECRET_ACCESS_KEY="your-secret-key"

PowerShell:

$env:MONGODB_URL = "mongodb+srv://<username>:<password>@cluster.mongodb.net/"
$env:AWS_ACCESS_KEY_ID = "your-access-key"
$env:AWS_SECRET_ACCESS_KEY = "your-secret-key"

4. Run the training pipeline

python demo.py

5. Start the application

python app.py

---

☁️ Cloud Setup

MongoDB Atlas

• Create a free M0 cluster
• Add 0.0.0.0/0 to Network Access for broad connectivity
• Push data to the cluster using notebook/mongoDB_demo.ipynb

AWS

• IAM: Create a user with AdministratorAccess and generate access keys
• S3: Create a bucket named my-model-mlopsproj (region: us-east-1) for model registry
• ECR: Create a repository named vehicleproj to store Docker images
• EC2: Launch an Ubuntu 24.04 t2.medium instance with 30GB storage; expose port 5080

---

🔄 CI/CD Pipeline

The pipeline is triggered on every push to the main branch via GitHub Actions.

Workflow:

1. Build Docker image
2. Push to AWS ECR
3. Pull and run on self-hosted EC2 runner

Required GitHub Secrets:

Secret	Description
AWS_ACCESS_KEY_ID	AWS IAM access key
AWS_SECRET_ACCESS_KEY	AWS IAM secret key
AWS_DEFAULT_REGION	us-east-1
ECR_REPO	ECR repository URI

---

🌐 Endpoints

Route	Description
/	Prediction UI
/training	Trigger model training

App is accessible at: http://<EC2-PUBLIC-IP>:5080

---

📊 Pipeline Components

Component	Responsibility
Data Ingestion	Fetches data from MongoDB, splits into train/test sets
Data Validation	Schema checks against schema.yaml
Data Transformation	Feature engineering, preprocessing
Model Trainer	Trains and evaluates candidate model
Model Evaluation	Compares against production model (threshold: 0.02)
Model Pusher	Pushes best model to AWS S3 registry

---

📝 License

This project is for educational and portfolio purposes.