Full-stack DevOps portfolio project built with Next.js 14, Prisma, and PostgreSQL. Dockerized with a multi-stage build, deployed to AWS EC2 via a GitHub Actions CI/CD pipeline (Docker Hub push → SSH deploy) that cut deploy time from 45 minutes to 3-5 minutes.

Full-stack SaaS-style fitness management platform for personal trainers. Features real-time chat, Google Calendar API integration, role-based authentication, dynamic workout planner, performance charts with E-1RM calculator, and meal plan management.

Full-stack e-commerce platform with separate Admin, Seller, and Buyer roles. Designed in Figma before implementation. Built with Node.js, Express and EJS templating with raw PostgreSQL queries. Features bcrypt authentication, product management, cart and order functionality.

Automated Number Plate Recognition system integrating YOLO object detection and PaddleOCR. Achieved 80% accuracy across diverse lighting and weather conditions with 20% improvement in data throughput over manual methods.

IEEE published research on satellite-based flood detection using deep learning. Trained a U-Net CNN on Sentinel-1 SAR imagery achieving 83% recall optimized for disaster response. Generated probabilistic severity maps for the Ganga floodplain.