At Sabudh Foundation, I analyzed real and synthetic images and benchmarked machine learning models for deepfake detection. I evaluated EfficientNet, SVM, and dual-stream CNN architectures, reaching up to 99.97% in-dataset accuracy, and found accuracy dropped to 50–53% under domain shift.
At AlgoUniversity, I built backend infrastructure for concurrent code compilation and execution on an online judge platform. I developed Docker-based execution sandboxes and asynchronous job pipelines, with Redis-based rate limiting and services deployed on AWS EC2.
