I am a junior at the School of Computer Science, Carnegie Mellon University.
I care deeply about ensuring that Artificial Intelligence benefits humanity. My research interests are in open-endedness, self-improving agents, automated discovery systems, and scalable oversight.
My hometown is Singapore. Bak chor mee, ramen, XO fish meehoon and beef rendang are some of my favorite foods. Outside of Computer Science, I enjoy photography (especially astrophotography), music (piano, guitar, beatboxing), badminton, and cooking :)
Personal Site | LinkedIn | Google Scholar
-
nanodiscover: Open-source reimplementation of TTT-Discover that replaces the Tinker dependency with local inference and training. Used by collaborators at the University of Minnesota for Evolution Fine-Tuning (co-author, Oral, ACM CAIS 2026 AIDWild Workshop) and referenced by the original TTT-Discover authors.
-
Search over Self-Edit Strategies for LLM Adaptation: First-author preprint. Search over self-generated training data and hyperparameters, allowing an LLM to invent its own adaptation procedure.
-
manual-code-canvas: Infinite-canvas tool for tracing unfamiliar codebases by hand, built because LLM Q&A was not enough to build real mental models.