Massimiliano (Max) Lupo Pasini obtained his Bachelor of Science and Master of Science in Mathematical Engineering at the Politecnico di Milano in Milan, Italy. The focus of his undergraduate and master studies was statistics and discretization techniques and reduction order models for partial differential equations. He obtained his PhD in Applied Mathematics at Emory University in Atlanta (GA) in May 2018. The main topic of his doctorate work was the development of efficient and resilient linear solvers for upcoming computing architectures moving towards exascale. Upon graduation, Max joined the Oak Ridge National Laboratory (ORNL) as a Postdoctoral Researcher Associate in the Scientific Computing Group at the National Center for Computational Sciences (NCCS).
In November 2019, Max became a computational scientist in the Scalable Algorithms and Coupled Physics Group in the Advanced Computing Methods for Engineered Systems Section of the Computational Sciences and Engineering Division at ORNL. In November 2020, Max became a data scientist in the Computational Coupled Physics group within the same section.
Max’s research focuses on the development of surrogate and generative AI models for material sciences, scalable hyper parameter optimization techniques for deep learning (DL) models, and acceleration of computational methods for physics applications. He was the technical lead of the Surrogates Models for Material Properties product within the Artificial Intelligence for Science and Discovery (AISD) thrust of the ORNL Artificial Intelligence Initiative in the fiscal years FY21, FY22, and FY23. He was the technical lead of the AISD thrust for fiscal years FY24, he is currently the technical lead of the project Scalable generative graph foundation models for atomistic materials modeling in FY25.
Presentation Title:
A Scalable Agentic Artificial Intelligence Scientific Workflow for Atomistic Materials Characterization
Presentation Abstract:
We present matsim-agents, an open-source agentic artificial intelligence (AI) workflow that combines surrogate and first-principles backends, portable execution across leadership-class high-performance computing (HPC) platforms, checkpoint-based fault tolerance, accuracy-driven escalation, and controlled cross-backend surrogate evaluation within a single framework. The matsim-agents workflow translates natural-language scientific objectives into typed, auditable atomistic-simulation plans and couples exploration based on machine learning interatomic potentials (MLIPs), selective first-principles labeling, model adaptation, and evidence-based escalation through a checkpointed driven orchestration layer driven by large language models (LLMs). Both surrogate and first-principles engines are configuration-selected backends, while machine-specific deployment is isolated from the workflow logic and checkpointed state enable execution across job failures and batch allocation boundaries. Controlled cross-backend evaluation reveals material-dependent outcomes, while measured surrogate accuracy determines whether the workflow continues the screening with MLIP models or escalates to DFT evaluation.