From file index to qualified alloy: Domino & Los Alamos National Lab- Paul Jojy & Gary Grider | Domino

Labs, agencies, and programs often hold decades of sensitive experiment and test records on HPC file systems and tape that nobody can search, with a rule that none of it leaves the enclave. Leveraging that trove of data into an AI workflow is complicated due to access and data movement restrictions, and every query takes a sysadmin and an afternoon.

This session walks through a materials discovery workflow in Domino that keeps every step inside the security boundary. GUFI (Grand Unified File Index), from Los Alamos National Lab, returns matching records on prior materials experiments in under a second across petabytes and respects permissions on the underlying files. Domino connects to the existing GUFI system and lets an agent search it. Matching records pull into a governed NetApp Volume, where a surrogate ML ensemble trained on that data generates and ranks new candidate alloys. A Slurm array job then runs hot isostatic pressing (HIP) simulations to qualify the candidates, and a qualification agent issues a verdict with a report.

The data access, HPC orchestration, model training and versioning, and the process verdict happen in a one governed, auditable environment. Domino deploys into your accredited environment, on-prem, air-gapped, or in cloud, and inherits its boundary. And the pattern fits any large archive of records you can’t move.