🚀 Student Performance Predictor: End-to-End MLOps on Google Cloud ☁️

Predicting student math scores with a production-ready, fully serverless ML system.

✨ Project Overview

This project delivers a complete MLOps pipeline, starting from raw data analysis to a live, scalable API endpoint on Google Cloud. The goal is to predict a student's math score based on key attributes like gender, parental education, and test preparation.

It serves as a full-stack demonstration of MLOps best practices, CI/CD automation, and cloud-native deployment.

---

🛠️ Tech Stack & Tools

Layer	Technologies Used
Data Processing	Python, Pandas, NumPy
Modeling	scikit-learn (Linear Regression, Random Forest)
Web Framework	Flask (REST API)
Containerization	Docker
Cloud Deployment	Google Cloud Run, Artifact Registry
Automation (CI/CD)	Google Cloud Build
Version Control	Git, GitHub
Monitoring & Logging	Cloud Logging, Docker logs

---

🎯 End-to-End Workflow

The entire ML system is structured into a seven-stage, production-ready pipeline:

1️⃣ Data Exploration & Model Building (Offline)

• Performed Exploratory Data Analysis (EDA) and deep-dive visualization in Jupyter.
• Conducted feature engineering: cleaned, analyzed relationships, handled missing values, and applied encoding/scaling.
• Trained and evaluated multiple ML algorithms (Linear Regression, Random Forest, etc.).
• Selected the best model based on the R² score and cross-validation, then serialized it using pickle.
  • Main Notebooks: notebook/EDA.ipynb, notebook/model_training.ipynb

2️⃣ Modular Pipeline Development (Code Refactoring)

The entire Jupyter workflow was refactored into a modular, reusable Python package to ensure reproducibility and maintainability.

• ✅ Implemented robust exception handling and logging across all modules.

3️⃣ Flask Web Application (API Layer)

Developed a lightweight Flask web application to serve the model for real-time predictions.

• HTML Form Endpoint: /predicted_data (Renders a form for manual user input).
• REST API Endpoint: /predict (Accepts JSON input, perfect for microservice communication).
• Includes input validation and structured error handling for a production environment.

4️⃣ Containerization with Docker 🐳

The entire application (model, dependencies, and Flask API) was containerized into a single, production-ready Docker image to guarantee environment consistency.

• Optimized Dockerfile for a smaller image footprint.
• Commands Used:

  docker build -t my-ml-api:dev .
  docker run -p 8080:8080 my-ml-api:dev

5️⃣ Deployment on Google Cloud Run ☁️

Deployed the containerized application to Google Cloud Run, a fully serverless platform, providing a managed, scalable REST API endpoint.

• Used Artifact Registry (ml-repo) to store and manage the Docker images.
• Deployed the service (my-ml-api) for autoscaling and zero infrastructure overhead.

6️⃣ CI/CD Automation with Cloud Build ⚙️

Implemented a complete Continuous Integration/Continuous Deployment (CI/CD) pipeline using Google Cloud Build and a GitHub trigger.

• Defined the CI/CD steps in cloudbuild.yaml.
• Every git push automatically triggers:
  1. Building the new Docker image.
  2. Pushing the image to Artifact Registry.
  3. Deploying the updated service to Cloud Run.

---

💡 Future Improvements (Roadmap)

• Vertex AI Integration: 🔁 Integrate Vertex AI Pipelines for automated, scheduled model retraining and advanced MLOps features.
• Observability: 📊 Add monitoring dashboards using Prometheus/Grafana or Cloud Monitoring to track model drift and API health.
• Testing: ✅ Implement automated unit tests for all modular components within the CI/CD pipeline.

---

👨‍💻 Author

Rakshith A H	ahrakshith122@gmail.com

⭐ If you found this project helpful, please give it a star!