Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Data Management topic
No spam. Unsubscribe anytime.
Panelists say data 'is infrastructure' for Genesys platform; metadata, provenance and conditioning are priorities
Summary
Data leads and attendees told DOE the platform must prioritize metadata, persistent identifiers, provenance and tools for conditioning heterogeneous research data; panelists promised a data-conditioning pipeline and working groups to refine requirements.
Panelists and attendees at the Genesys workshop emphasized that data — not just compute or models — is the critical bottleneck for agent-driven science. "Data is infrastructure," Kelly Rose said, urging awardees to plan for lineage, accreditation and transformation workflows to make data AI-ready. Participants raised three recurring needs: machine-readable metadata, persistent identifiers and automated transformation tools that preserve provenance while allowing cross-platform reuse.
Industry and academic contributors offered concrete ideas. Rohan Karthik (Inversion Semiconductor) recommended agentic tools to auto-tag and refactor large experimental outputs; Jeremy Moldavan (Cadence) suggested cryptographic linking and persistent IDs to allow discovery without exposing raw IP. "We have something called Jed AI platform that helps to deal with MPCs, their data... and how you manage it," Moldavan said, describing methods for traceability and verification.
Panelists described initial technical responses already in development: a memory layer to share skills and best practices, data cards for machine-readable metadata, agent-based auto-generation of metadata and a planned data-conditioning pipeline to prepare messy research outputs for model training. Kelly Rose asked teams to surface domain-specific needs so the platform team can prioritize tools and curation. The DOE team said catalogs on gear.doe.gov list current datasets and invite nominations for inclusion.
