I've built distributed backend inference services at Varcons Technologies, using Kubernetes and 16 GPU-accelerated Ray workers to process real-time streams from 600+ traffic cameras. I also developed a YOLOv8 detection pipeline sustaining 30+ FPS across those feeds.
At NVIDIA Graphics India, I built Python ETL and preprocessing pipelines that reduced batch preparation from eight hours to under five, trained supervised ML models that improved accuracy by 12% over baseline, and annotated 300+ egocentric video samples for action recognition. I also build full-stack and AI products including EduPlatform, TokenWise, and a RAG-based questionnaire-answering tool using Java, Spring Boot, React, FastAPI, LangChain, ChromaDB, Docker, and Kubernetes.
