Roshan Shetty

I’m a Data Scientist with 1+ year of combined experience across internships and contract work, focused on building production machine learning systems end to end. My work blends classical ML, MLOps, and cloud deployment, with a practical emphasis on automating data pipelines and reducing operational costs through applied AI.

In my Real-Time Fraud Detection System, I built a complete end-to-end fraud pipeline on the Kaggle Credit Card Fraud dataset—covering EDA, feature engineering, model training, deployment, monitoring, and streaming. I served real-time predictions via FastAPI with Pydantic validation and audit-ready storage in PostgreSQL, containerized the stack with Docker Compose, and automated CI/CD using GitHub Actions with fully automated AWS EC2 deployment.

I also strengthen reliability in production with data drift monitoring using Evidently AI and a streaming architecture powered by Redpanda (Kafka-API-compatible) to support both synchronous (FastAPI) and asynchronous fraud-checking workloads. Alongside building models, I’ve independently diagnosed and resolved infrastructure issues (Docker caching bugs, port conflicts, EC2 disk-space exhaustion, and CI/CD pipeline failures) and delivered additional ML/NLP systems like a multi-task customer support transcript analyzer.