Speaker
Description
Large-scale neutrino experiments increasingly depend on advanced computing technologies to support complex simulation, reconstruction, and analysis workflows. We present DUNE-GPT, a web-based computing and workflow platform developed as an enabling technology for the Deep Underground Neutrino Experiment (DUNE).
DUNE-GPT provides a unified interface for searching internal documentation and interacting with automated, scalable computing workflows that support physics-driven tasks. The platform integrates existing AI-enabled tools for document retrieval without AI model development, emphasizing robustness, usability, and application-level integration. DUNE-GPT has been deployed on world-class computing resources at the Argonne Leadership Computing Facility, including Polaris and Aurora, where initial production workflows were successfully executed using the Balsam workflow manager, validating scalable execution across heterogeneous architectures.
The platform supports workflows relevant to DUNE oscillation and neutrino interaction analyses by improving accessibility to institutional knowledge and automating large-scale computing tasks. Automated maintenance and document indexing pipelines on Fermilab-hosted systems enable sustained operation. This work highlights how modern computing platforms and workflow automation can directly enhance the efficiency and scalability of neutrino physics programs.