At IIT Kanpur, I developed a multi-stage YOLOv8 pipeline for detecting vehicle, helmet, and seatbelt violations across video streams. It achieved mAP@0.5 of 0.994 for bike/person and 0.937 for helmet/plate, with precision and recall above 0.92.
For SmartBot, a finalist project at the AWS × Battery Smart Hackathon, I built a real-time multilingual speech-to-speech support pipeline using FastAPI and Whisper STT. It combined LLM-driven intent classification with sentiment detection and a PostgreSQL-backed recommendation engine for nearby operational stations.
In an investor sentiment project, I processed StockTwits data with BERT-based models and integrated sentiment features into panel regressions studying abnormal stock returns. I’m an Economics undergraduate at IIT Kanpur, with experience spanning computer vision, NLP, and quantitative financial modeling.
