Speaker
Description
Modern heliophysics research is increasingly limited not only by model capability, but by fragmented workflows across literature, mission data archives, community scientific software, code, collaboration, and writing. Real research must connect physical interpretation with tools such as SPEDAS/PySPEDAS, mission data products, plots, manuscripts, and review processes. I argue that useful AI systems for scientific work should be designed as persistent, tool-using, auditable agents rather than one-off chat sessions.
Using workflows around Parker Solar Probe, SPEDAS/PySPEDAS, and AI-assisted scientific communication as motivating examples, I present lessons from building LingTai, a local-first prototype runtime for persistent AI research workers. The focus is not a single SPEDAS bot, but the broader agent-harness layer required for scientific work: project memory, tool execution, provenance, delegation, human approval, and reproducible artifacts. I discuss how such harnesses can support code navigation, data-to-plot reproduction, figure and artifact management, manuscript/review workflows, and community software maintenance while leaving scientific judgment with domain experts. I close with candidate evaluation tasks for measuring whether AI agents can reliably support real heliophysics research.