Nic Weber and Kyle Niemeyer
September 28, 2026
We are happy to announce a third cohort of the US Research Software Sustainability Institute (URSSI) Early-Career Fellowship. This cohort includes seven fellows working on the following projects:
Reliable Agentic Workflows for Sustainable Multi-Language Scientific Software Interfaces - Su Sun is a postdoctoral scientific software engineer in Chemical Engineering at Northeastern University. His project will investigate whether agentic AI workflows can reliably maintain and sustain cross-language interfaces for scientific software, using the Cantera MATLAB interface as a case study. You can read more at the project website.
From Pixels to Peaks: Sustainable Software for Global Seamount Discovery - Sarah Beethe is a PhD student in the College of Earth, Ocean, and Atmospheric Sciences at Oregon State University. Her project will rebuild the Seamount Catalog as open, Python-based infrastructure, then use that foundation to investigate the performance of machine-learning tools for identifying seamounts. You can follow her project here.
HDL-Verify: A Validation and Reproducibility Toolkit for AI Generated Hardware Description Code - Shima Mohaghegh is a graduate student at the University of Kansas where she works on computer architecture and hardware design in HPC. Her work is unique in that she is developing both a tool (HDL-Verify) and a benchmark for reference circuits based on the tool. You can read more about her work, and see her impressive progress so far, here.
Reliability Checks for Machine-Learning Surrogates in Scientific Software - Isaac Malsky works in computational astrophysics as Pipeline Scientist at the SETI Institute, supporting the TOLIMAN mission. His project will develop vibe-check, which in his own words addresses the following problem: “A surrogate (an emulator, neural operator, or learned solver component) is usually judged by its average test error. That number can hide problems: the model may be extrapolating outside its training data, the preprocessing may have leaked test data, or the outputs may break a physical constraint. vibe-check takes a predict function (or saved predictions) and your data splits, runs twelve checks, and writes a report that someone who did not train the model can read.” You can follow his work here.
Teaching Coding Agents the Brain Imaging Data Structure - Scott Huberty is a cognitive neuroscientist who currently works as a research fellow at the University of Southern California. His project will develop a prototype to help coding agents use BIDS (a community-developed specification, plus an ecosystem of tools, for organizing and describing neuroscience datasets).
Silent Failures in ML-Driven Simulations - Ali Mohaghegh is a PhD student in the Department of Aerospace Engineering at the University of Kansas. His project aims to produce an open-source diagnostic framework for detecting subtle reliability failures in scientific simulations that use machine learning (ML), reduced-order models (ROMs), neural operators, and other learned components. You can follow along here.
Navigating Successes and Failures in Agent-Assisted Code Development - Connor McClellan is a postdoctoral researcher in astrophysics at Michigan State University. His project will develop a new workshop module for the URSSI Schools on Research Software Development that asks what separates a successful integration of AI/ML coding tools from a failure. Participants will work through curated scenarios—solving similar problems both with and without agentic assistance—to build a concrete sense of when coding agents are likely to help, when they should not be used, and when restraints can counter their limitations. Because participants write code throughout, the workshop will also generate data on how agent usage affects code quality and maintainability, paired with exit and follow-up surveys on how it affects developers’ own understanding of the problems they solved.
We’re excited to bring these projects into the URSSI early-career program, and continue working with highly motivated scientists who take their software seriously.