pSCompose: Building A Web UI to compose perfSONAR Configurations in the Age of AI

Abstract

perfSONAR provides federated network measurement and is deployed at thousands of universities, research labs, and education networks, but turning it into repeatable measurement setups depends on pSConfig templates: JSON documents that link addresses, groups, tests, etc. Writing them by hand is error-prone and hard to reproduce across teams and sites, especially when they incorporate topology context and integrate with downstream scheduling. This poster presents pSCompose, a web-based UI for composing configurations that drive automated tests executed by perfSONAR test nodes. The application supports structured creation and management of measurement datatypes, including Templates, Tasks, Groups, Addresses, Tests, Schedules, Archives, and Contexts. By guiding users through the correct object relationships and generating the resulting pSConfig output, it aims to improve repeatability across teams and institutions. pSCompose is also a case study in building research software that the next maintainer can still fix, and a deliberate test of AI coding assistants. Development began around the time agentic coding tools such as Claude Code became available, and we used the project to find out where they help in a real software effort. The poster breaks down AI use across four stages (requirements, mockups and prototypes, software development, and packaging), highlights the intentional decisions made for maintainability, and closes with takeaways from building with AI. pSCompose has progressed from early mockups to prototypes and full deployments over the past year. It is now ready to ship, with packaging planned for inclusion in perfSONAR 5.3.0 by October. Next steps are usability testing and gathering requirements for the next release, and we invite feedback from the community.