Research Software Engineering at SNS and HFIR: Advancing Neutron Science Software in the Age of AI-Assisted Development

Abstract

The Spallation Neutron Source (SNS) and the High Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory are two of the most advanced neutron scattering facilities in the world, where thousands of researchers each year probe the structure and dynamics of materials at the atomic scale, advancing quantum materials, energy storage, medicine, soft matter, and engineering. The Neutron Data Project provides and maintains more than forty software packages supporting data reduction at all user program instruments (currently 31 instruments) across SNS and HFIR. This software is integral to hundreds of neutron scattering experiments each year, and the continuity of facility operations depends on its reliability. Our RSE team and RSE adjacent partners build and maintain this portfolio, spanning instrument-specific reduction codes, interactive visualization tools, web applications, and live data services, and also contribute to community analysis software. This talk presents an honest account of what it takes to build, modernize, and sustain a portfolio of this size in a facility environment. Every instrument has distinct reduction requirements, and highly reconfigurable instruments demand software that adapts to constantly changing instrument states. Several tools predate the careers of the engineers maintaining them, and legacy code must be modernized without disrupting an operating user program. A small team spread across many packages balances new development against maintenance, bug fixes, and beamline support tied to the facility run cycle. Behind the software sit persistent challenges: stabilizing builds and deployments across heterogeneous environments, taming dependency management, hardening CI/CD pipelines, chasing performance regressions, and capturing requirements from instrument scientists across diverse domains. We then describe how we are adapting this practice for the age of AI-assisted development. We have provisioned AI coding assistants for the RSE team and RSE-adjacent scientists and are piloting the use of AI to optimize selected phases of our software engineering lifecycle. We share our motivations, evaluation approach, and early experience, and commit to reporting back to the RSE community on what works, what doesn’t, and what we would advise others at similar scale. We will conclude with a summary of emerging best practices and recommendations for RSE teams adopting AI-assisted development.