Patrick Huck is a Senior Computing Engineer (CSE5) at Lawrence Berkeley National Laboratory (LBNL) with a Ph.D. in High Energy Nuclear Physics and 16+ years of experience acting as a vital bridge between complex scientific requirements and advanced computing infrastructure. As Co-PI and Principal Technical Architect of the Materials Project, he oversees delivery of AI-ready materials data to 779,000 researchers across 181 countries. As Lead Architect and Infrastructure Co-Lead for the American Science Cloud (AmSC) of the DOE’s Genesis Mission, he authored the platform’s foundational identity and architecture governance documents and leads AmSC’s Architecture Team. He is also Lead PI of ADEPT, a newly funded cross-divisional program at LBNL bridging experimental instrumentation and computational workflows between the Energy and Computational Sciences Areas.
Presentation Title (Co-Presenting with Debbie Bard):
Building the American Science Cloud: Federated Services, Zero-Trust Architecture, and Early Lessons
Presentation Abstract:
The American Science Cloud (AmSC) is building a secure, federated platform that connects AI models, scientific data, workflows, and computing resources across DOE laboratories, leadership computing facilities, experimental facilities, and commercial cloud providers. Its goal is to enable researchers to discover data and models, access large-scale computing, and execute AI-enabled workflows across institutional boundaries.
This talk will present both the emerging AmSC service ecosystem and the architecture that supports it. We will describe early capabilities including federated single sign-on, the Model Access Gateway, data and model services, common APIs, and tools for operating across distributed DOE resources. We will also introduce AmSC’s federated zero-trust approach, in which human, agent, and workload identities remain attributable across trust boundaries and access is governed through project-scoped credentials, policy enforcement, and auditable delegation.We will share lessons from integrating early science workflows and bringing independently developed services into a coherent platform. These experiences highlight that federation is not only an identity or networking problem: it also requires common interfaces, operational discipline, and close collaboration among facilities, service teams, and scientific users.