Agents for Scientific Computing

About

AI has evolved from generating text to operating as agents capable of retrieving data, executing code, and using tools. This talk explores how to apply these systems to automate tasks in scientific computing. As a practical example, I will outline the architecture of Archi, a retrieval agent currently deployed in CMS Computing Operations. Finally, we will address the engineering challenges of moving agents from read-only retrieval to active execution. This transition requires handling fragmented, conflicting, and stale data while mitigating the safety risks of autonomous actions. I will argue that strict guardrails and gradual deployment are necessary to reliably integrate these systems into high-energy physics workflows.

Bio:

Jason Mohoney is a Postdoctoral Associate in the Data Systems Group at MIT, working with Prof. Tim Kraska and Prof. Sam Madden. He completed his PhD at the University of Wisconsin-Madison, where his doctoral research focused on building cost-efficient and performant data systems, with a particular emphasis on data retrieval. He holds an undergraduate degree in Applied Math, Engineering, and Physics (AMEP) from Wisconsin. His current research focuses on AI agents, retrieval, and code generation, and he is a member of the Archi team, building an AI agent for CMS computing operations at CERN.

coffee will be served at 10h30

00:00:00 Slide 1

00:01:15 Slide 2

00:02:15 Slide 3

00:04:58 Slide 4

00:05:48 Slide 5

00:06:31 Slide 6

00:06:59 Slide 7

00:07:46 Slide 8