Jia Yu Lim — First-year PhD Student, Computer Science, National University of Singapore · WING Lab
First-year PhD Student, Computer Science
National University of Singapore · WING Lab
I am drawn to the problems nobody has cracked yet. What keeps me in AI is everything that happens after the benchmark, where a method either survives real data and makes something genuinely work better, or it quietly does not. Always up for a hard problem, a new sport, or a good conversation.
About
A little more
I am a first-year PhD student in computer science at the National University of Singapore, in the WING Lab, working on world models under the co-supervision of Prof. Min-Yen Kan and Prof. Robby Tantowi Tan.
Before this I read computer science with a major in artificial intelligence at Universiti Malaya, and worked on AI for finance and on graph neural networks including bank statement analytics for MSME credit scoring, and physics-informed graph networks for urban flood modelling.
What draws me to AI is the practical end of it: whether a method survives contact with real data, and whether the result reaches the community it was built for. I also like taking on things I am not yet good at including new problems, new sports, new anything. Trying is most of the point.
What I work on
World models
My current focus at WING Lab: models that build and use an internal representation of how a world behaves, rather than mapping inputs to outputs.
Multimodal document intelligence
Reading financial documents that were never meant to be machine-readable, retrieval, extraction and in-context modelling over statements and forms.
Graph neural networks for physical systems
Physics-informed graph networks for operational flood modelling, where the structure of the problem is as informative as the data.
Impactful AI
The thread through all of it: credit scoring for Malaysian MSMEs, urban flood forecasting, clinical risk prediction. I want to build things that change what someone can actually do, not only what a leaderboard says.
Latest work
UrbanFloodBench: Bridging AI and Hydrology through Benchmarking of Coupled 1D–2D Urban Flood Surrogate Models
Environmental Modelling & Software 2026
PRISM: Prompt-Refined In-Context System Modelling for Financial Retrieval
EMNLP 2026 Industry Track
Physics-Informed Graph Neural Networks for Operational Flood Modelling
IJCAI-ECAI 2026
When I am free
Please try not to book me outside office hours. I will almost certainly still be working, but let us both pretend otherwise. If a slot inside working hours looks clear, it is mostly fair game: send an invite to me and I will accept it.
Treat this as a rough guide rather than the truth. Plans move faster than calendars do, so confirm anything that matters by email first.