A journal paper in computer vision, a BSc thesis on detection for autonomous vehicles, and active work on AI for software engineering, where I study how to evaluate coding agents honestly.
A published study introducing an enhanced YOLOv9 model trained on the Multi-Weather Pothole Detection (MWPD) dataset for robust detection across adverse weather conditions.
A benchmark image dataset of road potholes captured across weather and lighting conditions, annotated with bounding boxes and released to support reproducible pothole detection research.
Submitted work on improving the factual consistency of large language model outputs by combining retrieval-augmented generation with knowledge graphs.
In-progress empirical study on whether process signals in a coding agent's trajectory, such as how messy its path looks, predict task failure under a strengthened evaluation.
In-progress research on building and, more importantly, measuring AI agents that carry out software engineering tasks honestly.
BSc thesis proposing a two-stage object detection framework for autonomous vehicles, built on a custom multi-weather dataset and YOLOv8.