Sudhanshu Shekar Kumar

I’m an AI/ML intern building applied intelligence that performs under real constraints—accuracy, latency, and actionable outputs. I combine strong ML foundations with production-minded engineering, from model training to inference pipelines and deployable APIs.

At DRC Systems Pvt. Ltd., I engineered an AI-powered Online Exam Proctoring System using a fine-tuned YOLO model (95% detection accuracy) to identify five violation classes: multiple faces, phone usage, abnormal head movement, speech, and browser violations. I implemented computer vision, audio analysis, and behavioral anomaly detection pipelines that generate real-time alerts, risk scores, and detailed integrity reports with evidence tracking, supporting inference at ~30 FPS for low-latency monitoring.

I also build agentic and explainable AI through my projects. For Artha Analytics, I designed an agentic AI platform for financial intelligence using LangGraph and the Groq API to orchestrate parallel multi-agent workflows for stock market analysis, structured report generation, and automated PDF exports via a Next.js and FastAPI architecture.

My work spans NLP and responsible decision support as well. I fine-tuned a T5 Transformer on a 26K+ Bitext Customer Support dataset to generate context-aware responses (BLEU score: 35) and deployed it via Hugging Face Hub and Spaces, and I created Sentiscope to benchmark sentiment models with SHAP explainability and deliver a REST API + web app using FastAPI and React.