Levi Harris

Hello! My name is Levi, I am a second-year masters student studying computer science at the University of North Carolina at Chapel Hill supervised by Dr. Tianlong Chen at the UNITES lab.

Today, my work centers on AI4Science, the task of applying artificial intelligence to established scientific fields. Recent breakthroughs (e.g., LLMs, image/video generation models, etc) offer a plethora of ways to improve a myriad of existing scientific processes. Now, disciplines as diverse as cell microbiology, algebraic topology, and political science are integrating AI-based techniques in unique ways to make life easier and do previously impossible things. And so, while AI is a remarkable, transformative technology, decades of cross-disciplinary work waits ahead of us to build effective, trustworthy systems together.

My research has focused on two, high-impact fields so far: metrology (the science of measurement) and meteorology (the science of weather).

AI4Metrology

• From late 2024 through spring 2025, researchers at Arizona State and I lead the development of novel, AI-techniques for 2D material image-acquisition and analysis. In march of this year, we published our findings in SPIE as SparseC-AFM: a deep learning method for fast and accurate characterization of MoS$_2$. Our key discovery here is that image super-resolution models can be adapted to predict full-resolution conductive maps from sparse inputs. With our proposed workflow, metrology researchers can image and analize 2D material samples in a fraction of the time.

AI4Meteorology

• Numerical Weather Prediction (NWP) models simulate the Earth’s atmosphere according to physical laws of nature. These prediction systems (e.g., ECMWF, GEFS) are the foundation of guidance for meteorologists, who use probabilistic data to generate discussions, forecast-products, and life-saving warnings for the general public. Predictably, NWP models are also some of the most computationally intensive applications ever devised by mankind. Each day, around the world, specially designed supercomputers crunch through the equations that predict future climate patterns. What if there was a way to shrink the compute required to simulate our Earth from the size of a warehouse down to your laptop? Moreover, what if this tiny climate-model actually outperformed the super-simulators that came before it? Today, AI is bringing this vision to reality.

• Researchers at the National Severe Storms Laboratory in Norman, Oklahoma recently began development of AI-WP models: the next frontier of weather prediction systems. Early results from NSSL’s WoFSCast and GenCast experiments offer a clear picture of where we’re heading – towards the development of rapidly-updating, high-resolution (i.e., <1km), CAMs with total coverage of the CONUS that are trusted and interogated by opperational meteorologists. In short, we imagine a world where weather modeling is no longer a bottleneck slowing the flow of life-saving weather forecasts to the public.

• AI4Meteorology is my current research focus. Specifically, I seek to develop trustworthy, AI-based techniques to support nowcasting of short-term (i.e., 0-6H), high-impact weather events.

Bonus!

AI4Sports

During my undergrad at UNC, I also worked with Dr. Gedas Bertasius to build densely annotated datasets for structured video generation and human behavioral modeling using basketball broadcast footage.