The University of North Florida has received a grant from the Alfred P. Sloan Foundation to build generative AI tools that speed up and improve the reliability of updating research software code. The award, announced September 9, marks the first Sloan Foundation grant to UNF.
The project targets a persistent problem in scientific computing: legacy code that underpins research often becomes difficult to maintain as scientific needs change. "The durability and trustworthiness of scientific research depends on the rigor and robustness with which the underlying research code is built and maintained," said Joshua M. Greenberg, director of the Sloan Foundation's Technology program. "This grant to UNF will help ensure that we can take advantage of new AI tools to safely restructure legacy research software as scientific needs evolve."
Who is leading the work
The research is led by School of Computing faculty Dr. Upulee Kanewala, associate professor, and Dr. Nan Niu, director and professor. Two graduate research assistants, Eric Good and Nabin Chaulagain, are working on the project.
The team will evaluate generative code approaches for refactoring - changing the internal structure of legacy computer code to make it cleaner and easier to maintain without altering how it functions or affecting the precision of calculations. The goal is to help researchers across the globe evolve their software while preserving accuracy and reliability.
Testing for correctness
A key component of the work is pairing generative AI-supported refactoring with metamorphic testing, a technique for verifying that software behaves correctly when there is no simple oracle to check outputs against. The project aims to provide evidence that this combination can reduce technical barriers to maintaining and extending research software.
Expected outcomes include increased confidence in refactoring legacy scientific code, improved software engineering methods for developers working under resource constraints, and a foundation for larger-scale research focused on sustainable research software development. By reducing the effort needed to update research codes, the team hopes to cut the time researchers spend managing software.
Why this matters for science and research professionals
Researchers who maintain their own analysis pipelines, simulation code, or data processing scripts know how much time legacy code can consume. This project is building evidence for whether AI for science and research workflows can handle refactoring tasks without introducing subtle numerical errors - the kind that can invalidate results. If the approach holds up, it could reduce the maintenance burden that currently competes with actual research time.
Your membership also unlocks:
