Published Paper Details:

The need for effective, reliable, and accurate screening methods to assist immigration authorities has been highlighted by the rise in US work visa applications in recent years. In order to predict the approval status of US visa applications, this project offers a thorough end-to-end Machine Learning Operations (MLOps) pipeline that makes use of employer and applicant data. The method uses important characteristics like education level, work experience, company size, geography, and prevailing wage to handle binary classification--certified or refused. To guarantee excellent data quality and pipeline consistency, preprocessing methods include transformation pipelines, schema validation, and Evidently AI data drift detection. In order to ensure that only the best-performing model is pushed to production, the deployment pipeline integrates GitHub Actions for Continuous Integration and Continuous Deployment (CI/CD), Docker for containerization and reproducibility, and AWS services (EC2 and ECR) for hosting and model storage. The architecture encompasses the entire MLOps lifecycle, including data ingestion from MongoDB Atlas, validation, transformation, model training with hyperparameter tuning, performance evaluation, model registry, and deployment. A user-friendly web interface developed with Streamlit and FastAPI supports real-time prediction and provides stakeholders with instant feedback. With a clear folder organization, configuration-driven design, virtual environment separation, and smooth automation, the system is expandable and modular. Important engineering concepts including model versioning, robust error handling, logging, and dynamic environment variable management are highlighted in this project. Finally, it illustrates how MLOps can improve machine learning systems' scalability, dependability, and transparency in crucial decision-making areas like immigration and law enforcement.