Mostly, I make slow and complicated things fast and usable.

We make physics simulation autonomous. Benchmarked against the major neural operator architectures on NS2D, our model set a new state of the art for physics-informed machine learning; the full benchmark table is here. Our simulations run about 1,000 times faster than traditional CFD. Reynolds turns a plain-English description into a complete, validated simulation. We raised USD 2.4M from Y Combinator, Square Peg, and Great Wave. Our work now spans Kajima, the Government of India, GovTech Singapore, government infrastructure, and smart buildings.

Built physics-informed storm surge forecasting for operational water decisions. It ran 600 times faster, improved accuracy by 10%, and doubled the useful forecast horizon.

Implemented Physics-Informed Point Networks in the Julia scientific computing ecosystem, so PDEs on irregular geometries became a package rather than a research prototype. I was the programme's only high school student that year.

Worked on the flight computer and power system for a satellite built by high school students. It launched on a SpaceX Falcon 9 into a 510 km orbit and is still operational. I also fit image compression and onboard image selection into 512 KB of memory.

Added parallel and GPU computing to quantum simulations with A*STAR IHPC. The result was about 100 times faster and much easier for other researchers to use.

Built object detection, simulation, and physics-based reinforcement learning systems for robots, then helped make the physical prototypes work too.

Built a flexible pressure sensor for medical monitoring. The project went on to win Third Grand Prize in Biomedical Engineering at ISEF.