India
Building production-ready deep learning pipelines for automated civic infrastructure monitoring at Airawat Research Foundation, IIT Kanpur's Technopark , directly extending my M.Tech thesis research at IIT Kanpur Key contributions: ▸ Trained and delivered YOLOv12 Medium detection models for 4 civic domains — road damage, garbage overflow, illegal hoardings, and fallen trees — achieving F1-scores up to 87.6% and mAP@50 up to 0.88. ▸ Built a full model development pipeline: 31K+ annotated images, frame extraction, augmentation, class balancing, hard negative mining, and iterative retraining from real-world field testing feedback. ▸ Developed a Flask + SQLite web app for false positive management, enabling structured detection review and automated retraining dataset export — reducing false positive rates measurably across every cycle. ▸ Benchmarked YOLOv8 through YOLOv12 across multiple model sizes; identified YOLOv12 Medium as optimal for precision, localisation accuracy, and inference speed.
The course will skill students on effective research communication as well as introduce them to some commonly used research methodologies and paper-writing techniques used in various sub-areas of computer science including theory, systems and data science.