I am currently an AI Engineer at CtrlCV, where my primary work focuses on the full-stack development of a road defect management application. The computer-vision-based application analyses dashcam footage to detect road assets and defects for maintenance tracking.
While my current responsibilities focus on full-stack engineering, my primary interests lie in the realm of computer vision and machine learning. I enjoy computer vision because of the interesting paradox of visual data: humans can easily classify everyday objects where machines struggle, yet they can distinguish micro-categories at a scale that would overwhelm human capability. For my dissertation, I developed a synthetic urban image data generation pipeline in Unreal Engine 5 with automated ground-truth annotation to improve urban segmentation accuracy. I am actively seeking dedicated roles in these fields where I can translate state-of-the-art computer vision research to real-world applications.
Supervised by Professor Andrew French, I built a synthetic data generation pipeline in Unreal Engine 5 to explore how synthetic data can improve model performance.The pipeline first traverses the 3D world to capture standard images, then procedurally modifies scene assets to their respective ground-truth materials to capture the corresponding ground-truth images.By integrating the synthetic data with real data to train a U-Net model on my university's remote GPU cluster, I improved the average class IoU from 0.578 (real data alone) to 0.595.
Project Niukka is a local-first personal assistant web application that I use almost daily. I built it with the intention of consolidating the tools I find helpful into a single place. The home page features a minimal canvas for jotting down random ideas, alongside a rotating photo slideshow to bring a personal touch and keep good memories in view. I also extended this canvas concept to implement a task management system inspired by the book Four Thousand Weeks. This three-canvas layout restricts focus to three active tasks, providing a dedicated, free-form writing space to jot down anything about each task at hand.
One of the main drivers for this project was to have a financial ledger tailored specifically for my needs. This was also a good way to hone my full-stack development skills while working on a fun personal project.Beyond the financial tools, the system includes a personal journal, a task calendar, and a Spanish-vocabulary song lyric quiz. I am planning to integrate computer vision to simplify ledger entries and fine-tune local LLMs once the basic functionalities are complete.
I led a three-person team, including Chen Ing Low and Gabriel Yong, to develop an OSA prediction webpage to facilitate OSA diagnosis.Using raw clinical datasets collected from Malaysian hospitals, I experimented with machine learning methods to maximize classification accuracy. Evaluated with leave-one-out cross-validation due to dataset size concerns, our random forest model scored 0.834 in classifying OSA presence and 0.515 in grading its severity.
The best way to contact me is through email. Feel free to reach out to me regarding any opportunities or small-scale projects within my area of expertise. I would be happy to discuss your ideas.
