MLOps Loan Approval Prediction System

Overview

This project implements a production-grade MLOps system that predicts loan approval outcomes based on applicant data. It integrates machine learning with modern DevOps practices, covering model training, versioning, monitoring, and deployment. The system is fully containerized and deployed on Google Cloud Run with automated CI/CD via GitHub Actions.

Background

Manual loan approval is inherently slow and costly due to repetitive human tasks and complex paperwork, which delays decisions for customers.

This project demonstrates how machine learning can automate and standardize such decisions, backed by an end-to-end MLOps workflow: data versioning, model training, experiment tracking, monitoring, CI/CD, and deployment on Google Cloud Run.

Constraints

Due to the priority scale, I haven't set up and run testing for model, data, or API components. The monitoring alert was also only for system metrics.

Key Features

• ✅️ Microservices architecture (Frontend, API, Model, Monitoring)
• ✅ Data versioning with DVC + DagsHub remote storage
• ✅ Model tracking with MLflow + DagsHub Registry
• ✅ Drift detection using Evidently
• ✅ Containerized services with Docker & Docker Compose
• ✅ CI/CD with GitHub Actions, Deploy using Cloud Run
• ✅ Structured logs for prediction audit trail

Methodology

It begins with preparing loan data by cleaning and preprocessing it. Then models were built, utilizing MLflow for tracking and stored into a DagsHub repository. A model that is ready for production (use in model service) will then be registered.

The system separates the model, API frontend, and monitoring services into different services. These services are first built and tested thoroughly on a local machine to ensure all components function correctly. Once ready, GitHub Actions automates the process for the whole service: the Docker images are built, then pushed to Google Cloud Artifact Registry, and finally deployed to Cloud Run.

The prediction results (structured logs) from the API service will be logged automatically with Cloud Logging, then sunk to BigQuery. The monitoring service (Evidently) will use these logs to detect data drift, and the report is updated daily. For the system metrics, it utilizes GCP's native Cloud Monitoring, which can also send alerts.

Potential Impact

• Faster, Cheaper Operations: The system cuts approval times from days to just minutes or seconds, significantly lowering the bank's operational costs by automating time-consuming manual work.
• Better Risk Decisions and Fairness: Machine learning provides a much more accurate assessment of risk, but the implementation process also requires the bank to thoroughly evaluate the system for fairness and ethical bias to ensure compliance and equitable outcomes.
• Reliable and Stable Deployment: MLOps ensures the system is stable and continuously monitored for technical issues and data problems, guaranteeing the reliability of loan decisions and allowing for quick, safe updates to be deployed.

Conclusion

This project demonstrates how MLOps principles can bridge the gap between model development and real-world deployment. It automates deployment and monitoring while maintaining reproducibility and scalability. Through this build, I gained deep hands-on experience in cloud infrastructure, workflow automation, and end-to-end system design.