Technical Demonstrations at the DOE Booth

This demonstration explores using natural language to run ICF simulations on LLNL’s HPC and codes.

In this demo a crew of AI agents configure, start, monitor and control the 3D printing of a metal part. Using a new agent-in-the-loop control methodology, this demo showcases how researchers can inject human intuition into control decisions and make on-the-fly changes to control logic in response to new information. The agents use a combination of human guidance, in-situ monitoring data and simulations to make time-sensitive decisions regarding the printer parameters for each layer to control the material properties of the metal part. A human operator manages the workflow from a chat interface, seamlessly directing a workflow that bridges the cloud (AI agents), edge (3D printer) and the Oak Ridge Leadership Computing Facility (simulations).

This demo will operate with a digital twin of the 3D printer in the place of a real printer. There will be videos and printed parts of the system being used to print real parts at Oak Ridge National Laboratory’s Manufacturing Demonstration Facility.

The demo will consist of a series of interactive (simulated) prints, with visuals of the chat interface, the digital twin simulation of the print with synthetic infrared camera feeds, the results of forward-looking simulations to inform decisions and the reasoning of the agents making control decisions.

This demo will cover the new, open-source SNL Application Data Collection Tools and how the practical benefits it brings to AI may reduce time to result for HPC workloads.

Systems performing scientific computing, data analysis and machine learning tasks have a growing demand for application-specific accelerators that can provide high computational performance while meeting strict size and power requirements.

However, the algorithms and applications that need to be accelerated are evolving at a rate that is incompatible with manual design processes based on hardware description languages. Agile hardware design tools based on compiler techniques can address these limitations by quickly producing an application-specific integrated circuit (ASIC) accelerator starting from a high-level algorithmic description.

In this talk, researchers present the software-defined accelerator (SODA) synthesizer, a modular and open-source hardware compiler that provides automated end-to-end synthesis from high-level software frameworks to ASIC implementation, relying on multilevel representations to progressively lower and optimize the input code. Our approach does not require the application developer to write any register-transfer level code, and it is able to reach up to 364 GFLOPS/W on typical convolutional neural network operators.

The Consortium for the Advancement of Scientific Software (CASS) is a federation of six software stewardship organizations (SSOs) funded by the U.S. Department of Energy (DOE) Office of Advanced Scientific Computing Research (ASCR) to help steward DOE/ASCR software products, most of which were developed or enhanced in the Exascale Computing Project and continue to be widely used in computational science and engineering projects on DOE and other high-performance computers. This demonstration will provide an overview of CASS, its member organizations, the software portfolio and updates on recent activities.

OLCF’s Slate platform is designed to support Kubernetes workloads and enables researchers to self-manage their open and moderate enclave deployments such as databases, message brokers, web portals, etc., through a low barrier web interface.

The talk will conclude with a demonstration of an end-to-end workflow that includes launching a database from the Slate web portal, data generation on Frontier, OLCF filesystem access from Slate’s JupyterHub platform, data ingestion into the database, and viewing that data from a user friendly front-end or via traditional queries.

A demonstration of end-to-end execution of the initial prototype. The prototype shows the first end-to-end system that coupled 5G-connected agricultural sensors with HPC simulations.

Live telemetry from the Citrus Under Protective Screen (CUPS) facility was streamed through a private 5G network, processed by FabriStack and Laminar, and used to trigger OpenFOAM CFD simulations on HPC clusters at UCSB and Notre Dame. The system demonstrated real-time data-driven simulation, adaptive resource scheduling with pilot jobs and multi-site portability — validating the vision of a digital-physical fabric that unifies sensing, wireless networking and large-scale computing for near real-time agricultural decision support.

This demo showcases the power and flexibility of ANARI, the Khronos standard for portable, high-performance rendering. We will demonstrate how ANARI enables scientists and developers to write visualization code once and run it efficiently on a variety of rendering backends, from CPUs to GPUs, without changing their application. The session will feature live examples running on supercomputing resources, highlighting performance portability, scalability, and ease of integration into existing HPC workflows.

ExaChem is a suite of scalable electronic structure methods to perform ground and excited-state calculations on molecular systems. The methodologies in ExaChem are implemented using the Tensor Algebra for Many-body Methods (TAMM) library. TAMM is a parallel tensor algebra library for performance-portable development of scalable electronic structure methods that can be run on modern exascale computing platforms. ExaChem and TAMM are actively being developed and maintained at the Pacific Northwest National Laboratory (PNNL) and distributed as open-source under the terms of the Apache License version 2.0.

Advances in AI, machine learning and data generation are changing the data and computing landscape in biotechnology rapidly. The Data Ecosystem infrastructure being developed by the DOE SC Office of Biological and Environmental Research is enabling data exploration, modeling and inquiry at scale across the DOE HPC ecosystem via the JGI Analysis Workflow Service (JAWS) supporting Joint Genome Institute (JGI), Environmental Molecular Sciences Laboratory (EMSL) and the BER-ASCR OPAL AI and lab automation project.

This demonstration will explore how the JGI distributes genomic analyses across DOE computing facilities on behalf of its users without their intervention, opening opportunities for larger-scale analysis and greater resilience. JAWS demonstrates how the DOE Integrated Research Infrastructure (IRI) vision can be realized in practice. JAWS is a multi-site workflow execution system that orchestrates large-scale, high-throughput genomic workflows across the ASCR computing facilities and the DOE HPC infrastructure, including NERSC (Perlmutter), LBNL (Dori and Lawrencium), EMSL (Tahoma), ALCF (Crux) and OLCF (Defiant), while ensuring secure, automated data movement via Globus and portable execution through containers.

By integrating community workflow standards with DOE’s HPC ecosystems, JAWS provides a scalable, fault-tolerant backbone for diverse scientific campaigns, from metagenome analysis in the National Microbiome Data Collaborative (NMDC) to multi-omics integration for carbon cycling studies. Within the IRI Testbed, JAWS showcases how interoperable services can bridge experimental, observational and computational facilities, enabling resilient cross-facility workflows, accelerating feedback between instruments and simulations and empowering researchers with FAIR, AI-ready data products. This demonstration will highlight both JAWS’ technical capabilities and the science it enables.

The National Environmental Policy Act (NEPA) requires federal agencies to conduct environmental reviews for proposed major federal actions disclosing the potential environmental impact of actions. This process has been generating a vast amount of review documents that contain rich information and potential to inform and streamline future review process. However, the document repositories are scattered across multiple databases and organizations. Furthermore, these reviews typically require enormous amounts of human labor and time, creating significant bottlenecks that can delay project completion by months or even years. To address these challenges, the PermitAI project is developing a suite of AI web applications to streamline environmental permitting through innovative technology.

This demo will showcase SearchNEPA and CommentNEPA, the two flagship research prototype solutions from the PermitAI suite which are currently being tested by several federal users in beta mode.

SearchNEPA provides advanced discovery capabilities across approximately 140,000 NEPA documents, including Environmental Impact Statements, Environmental Assessments and Categorical Exclusions. The platform features customizable filters, extensive metadata and integrated chatbot functionality that allows users to quickly retrieve information from this massive database.

CommentNEPA employs a Human-AI agentic framework to autonomously process public correspondence letters, delineate comments, extract concerns and categorize them systematically into bins. The system incorporates AI and human feedback loops to continuously optimize AI prompts and improve generated summaries and concern extraction accuracy.

Both tools support the U.S. Department of Energy’s mission to convert vast data repositories into AI-ready resources and accelerate energy infrastructure deployment by streamlining permitting processes with AI technology. These applications aim to help reduce project timelines while maintaining thorough environmental oversight.

A key capability of the anticipated Integrated Research Infrastructure (IRI) is to run compute jobs across distributed facilities within the lifecycle of the same application workflow. Data storage, access and movement must be tightly integrated with job execution within this workflow.

This demo will present an overview of an IRI agent that integrates with example resource providers, including HTCondor (compute job meta-scheduler), SENSE (cross-facility networking), Janus (data movement), EJFAT (data streaming and processing) and an example common interface for placing, moving and executing jobs over diverse facilities.

Following the overview, the EJFAT team (ESnet/Jefferson Lab FPGA Accelerated Transport) will show examples of real time streaming of raw instrument data from Berkeley Lab’s Advanced Light Source and Argonne’s Advanced Photon Source (APS).

This demo showcases a proof-of-concept IRI workflow where compute jobs are placed based on available resources across facilities, and datasets are moved and accessed via on-demand network and data transfer services. This demonstration will discuss how future IRI agents and interfaces will help orchestrate such workflows with general resource providers.

ASCRIBE-VR enables researchers to interactively explore and analyze complex scientific datasets in a Virtual Reality environment, enhancing understanding of AI outputs and accelerating discovery from 3D image sets. The system integrates advanced visualization techniques with user-friendly interfaces for seamless data analysis.

The Integrated Research Infrastructure Pathfinder Projects advance the IRI program by implementing multi-facility workflows in a way that provides actionable information for other projects and the program as a whole. This demo will showcase progress over the past year by showing the ability for multiple light-source beamlines to run at multiple HPC facilities. This demonstration will show this for beamlines at three light source facilities — the ALS, LCLS and APS — each able to run at multiple HPC facilities.

Advances in AI, machine learning and data generation are changing the data and computing landscape in biotechnology rapidly. The Data Ecosystem infrastructure being developed by the DOE SC Office of Biological and Environmental Research is enabling data exploration, modeling and inquiry at scale across the DOE HPC ecosystem via the JGI Analysis Workflow Service (JAWS) supporting Joint Genome Institute (JGI), Environmental Molecular Sciences Laboratory (EMSL), and the BER-ASCR OPAL AI and lab automation project.

This demonstration will explore how the JGI distributes genomic analyses across DOE computing facilities on behalf of its users without their intervention, opening opportunities for larger-scale analysis and greater resilience. JAWS demonstrates how the DOE Integrated Research Infrastructure (IRI) vision can be realized in practice. JAWS is a multi-site workflow execution system that orchestrates large-scale, high-throughput genomic workflows across the ASCR computing facilities and the DOE HPC infrastructure, including NERSC (Perlmutter), LBNL (Dori and Lawrencium), EMSL (Tahoma), ALCF (Crux), and OLCF (Defiant), while ensuring secure, automated data movement via Globus and portable execution through containers.

By integrating community workflow standards with DOE’s HPC ecosystems, JAWS provides a scalable, fault-tolerant backbone for diverse scientific campaigns, from metagenome analysis in the National Microbiome Data Collaborative (NMDC) to multi-omics integration for carbon cycling studies. Within the IRI Testbed, JAWS showcases how interoperable services can bridge experimental, observational and computational facilities, enabling resilient cross-facility workflows, accelerating feedback between instruments and simulations, and empowering researchers with FAIR, AI-ready data products. This demonstration will highlight both JAWS’ technical capabilities and the science it enables.

NexGen Analytics (NGA), in collaboration with Sandia National Laboratories (SNL), presents WorkVisualizer, a novel performance profiling and visualization tool for accelerating scientific and AI/ML applications.

As HPC developers adopt heterogeneous architectures to advance their codebases, the potential performance benefits are often offset by the added complexity in developing and optimizing portable code. Developers at SNL and other DOE laboratories have struggled with profilers that offer valuable insights into program execution but require sifting through large amounts of data or complex traces to isolate problem areas in their application’s code. To address these challenges and enable WorkVisualizer to provide quick, actionable insights, NGA has collaborated with the OVIS/LDMS team from SNL to implement scalable data collection and predictive, AI-driven analysis capabilities. In doing so, WorkVisualizer can offer a high-level view of a user’s code that clearly highlights anomalies, bottlenecks and other performance issues.

This demonstration will show how the prototype of WorkVisualizer can be used to pinpoint a performance issue in ExaMiniMD. Then, researchers will discuss the improvements that have come about as a result of our collaboration with the OVIS/LDMS team. Finally, researchers will preview the new features and capabilities that will be added in the next year.

The WorkVisualizer is supported by the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research through the Small Business Innovation Research (SBIR) Program, under SBIR Phase II Award DE-SC-0024832.

A key capability of the anticipated Integrated Research Infrastructure (IRI) is to run compute jobs across distributed facilities within the lifecycle of the same application workflow. Data storage, access and movement must be tightly integrated with job execution within this workflow.

This demo will present an overview of an IRI agent that integrates with example resource providers, including HTCondor (compute job meta-scheduler), SENSE (cross-facility networking), Janus (data movement), EJFAT (data streaming and processing) and an example common interface for placing, moving and executing jobs over diverse facilities.

Following the overview, the EJFAT team (ESnet/Jefferson Lab FPGA Accelerated Transport) will show examples of real time streaming of raw instrument data from Berkeley Lab’s Advanced Light Source and Argonne’s Advanced Photon Source (APS).

This demo showcases a proof-of-concept IRI workflow where compute jobs are placed based on available resources across facilities, and datasets are moved and accessed via on-demand network and data transfer services. We will discuss how future IRI agents and interfaces will help orchestrate such workflows with general resource providers.

In this demo a crew of AI agents configure, start, monitor and control the 3D printing of a metal part. Using a new agent-in-the-loop control methodology, this demo showcases how researchers can inject human intuition into control decisions and make on-the-fly changes to control logic in response to new information. The agents use a combination of human guidance, in-situ monitoring data and simulations to make time-sensitive decisions regarding the printer parameters for each layer to control the material properties of the metal part. A human operator manages the workflow from a chat interface, seamlessly directing a workflow that bridges the cloud (AI agents), edge (3D printer) and the Oak Ridge Leadership Computing Facility (simulations).

This demo will operate with a digital twin of the 3D printer in the place of a real printer, but will have videos and printed parts of the system being used to print real parts at Oak Ridge National Laboratory’s Manufacturing Demonstration Facility.

The demo will consist of a series of interactive (simulated) prints with visuals of the chat interface, the digital twin simulation of the print with synthetic infrared camera feeds, the results of forward-looking simulations to inform decisions and the reasoning of the agents making control decisions.

The Facility for Rare Isotope Beams (FRIB) is a DOE Office of Science user facility focused on studying problems of national interest in low-energy nuclear physics. Real-time or near real-time (nearline) analysis methods are critical tools for enabling FRIB science as new detectors and data acquisition technologies which allow for higher data rates and volumes are incorporated into laboratory systems. In addition, many experiments rely on computationally intensive analysis tasks where real-time processing on local computer systems is not feasible.

EJFAT/E2SAR provides a convenient platform for streaming data from FRIB to offsite locations, for example, to NERSC, where more computational resources are available to experimenters. An automated workflow was developed to stream data from an FRIB experiment to NERSC, process digitized detector waveform traces at NERSC, extract features of interest from the trace data and send the results back to FRIB.

This demonstration shows an example of how EJFAT/E2SAR can be used in future applications to connect the data stream from a live experiment to external high-performance computing resources.

Inspired by principles of the brain, SpiNNaker2 is a many-core neuromorphic chip designed for large-scale asynchronous processing. The flexibility provided by its reconfigurability, scalability afforded by its real-time, large-scale mesh and native support for hybrid acceleration of symbolic spiking and deep neural networks make SpiNNaker2 a unique computing platform. Sandia National Laboratories has partnered with SpiNNcloud to explore computational advantages neuromorphic computing can enable for a variety of applications. This demo will showcase the SpiNNaker2 architecture across a range of applications – large language models (LLMs), scientific computing, optimizations and AI/ML.

LANL has been exploring 3D technologies to help advance scientific understanding including VR headsets, 3D monitors, holographic displays and more. This demonstration will showcase a sampling of the technologies LANL researchers use and how they lead to new insights in big datasets.

E4S is a curated, Spack based software distribution of 100+ HPC and AI/ML packages. E4S, an HPSF project, is a platform for product integration and deployment of applications such as Quantum Espresso, OpenFOAM, LAMMPS, performance evaluation tools such as TAU, HPCToolkit, DyninstAPI, PAPI, numerical libraries such as Trilinos, PETSc, runtime systems such as Kokkos, MPICH, OpenMPI, Raja, and supports both bare-metal and containerized deployment for CPU and GPU platforms.

E4S provides a Spack binary cache and a set of base and full-featured container images with vendor runtimes to support GPU architectures from NVIDIA, Intel, and AMD. E4S is a community effort to provide open-source software packages for developing, deploying and running scientific applications and tools on HPC platforms. It has built a comprehensive, coherent software stack that enables application developers to productively develop highly parallel applications that effectively target diverse exascale architectures. ParaTools Pro for E4S(TM) supports commercial cloud platforms, including AWS, GCP, Azure and OCI and supports Torque and SLURM schedulers.

The Consortium for the Advancement of Scientific Software (CASS) is a federation of six software stewardship organizations (SSOs) funded by the U.S. Department of Energy (DOE) Office of Advanced Scientific Computing Research (ASCR) to help steward DOE/ASCR software products, most of which were developed or enhanced in the Exascale Computing Project and continue to be widely used in computational science and engineering projects on DOE and other high-performance computers. This demonstration will provide an overview of CASS, its member organizations, the software portfolio and updates on recent activities.

Join Robert Rallo for an insightful talk on how a new AI-powered agent-based platform integrates various AI models to expedite chemical discovery, particularly in catalysis, making research more efficient and vastly increasing the pace of innovation. This innovative platform, described as a “co-scientist,” empowers researchers to simulate and explore 3-D molecular structures, analyze data and translate validation protocols into actionable instructions for robotic laboratory equipment, enhancing both the speed and precision of experiments. Don’t miss this opportunity to learn how new technology is reshaping the future of scientific exploration.

Experience millisecond-resolution network monitoring in action! This demonstration shows how Linux systems can capture and report fine-grained telemetry in real time, enabling HPC and AI workloads to achieve unprecedented visibility into network performance.

This demonstration will cover the new, open-source SNL Application Data Collection Tools and how the practical benefits it brings to AI may reduce time to result for HPC workloads.

This demo will provide highlights of a post-workshop report: Toward Next-generation Ecosystems for Scientific Computing: Harnessing Community, Software, and AI for Cross-Disciplinary Team Science. The report is an outcome of a workshop held on April 29–May 1, 2025.

The event brought together more than 40 experts (including staff from ANL, LBNL, LLNL, LANL, ORNL, PNNL, SNL) from HPC, AI, computational science, software engineering, applied mathematics, social sciences and community development to chart a path toward more powerful, sustainable and collaborative scientific software ecosystems.

Report of the 2025 Workshop on Next-Generation Ecosystems for Scientific Computing: Harnessing Community, Software, and AI for Cross-Disciplinary Team Science. L.C. McInnes, D. Arnold, P. Balaprakash, M. Bernhardt, B. Cerny, A. Dubey, R. Giles, D.W. Hood, M.A. Leung, V. Lopez-Marrero, P. Messina, O.B. Newton, C. Oehmen, S.M. Wild, J. Willenbring, L. Woodley, T. Baylis, D.E. Bernholdt, C. Camano, J. Cohoon, C. Ferenbaugh, S.M. Fiore, S. Gesing, D. Gomez-Zara, J. Howison, T. Islam, D. Kepczynski, C. Lively, H. Menon, B. Messer, M. Ngom, U. Paliath, M.E. Papka, I. Qualters, E.M. Raybourn, K. Riley, P. Rodriguez, D. Rouson, M. Schwalbe, S.K. Seal, O. Surer, V. Taylor, and L. Wu. Report ANL-25/47, 2025, https://doi.org/10.48550/arXiv.2510.03413.

The Integrated Research Infrastructure Pathfinder Projects advance the IRI program by implementing multi-facility workflows in a way that provides actionable information for other projects and the program as a whole. This demonstration will showcase progress over the past year by showing the ability for multiple light-source beamlines to run at multiple HPC facilities. We will show this for beamlines at three light source facilities — the ALS, LCLS and APS — each able to run at multiple HPC facilities.

A demonstration of end-to-end execution of the initial prototype. The prototype shows the first end-to-end system that coupled 5G-connected agricultural sensors with HPC simulations. Live telemetry from the Citrus Under Protective Screen (CUPS) facility was streamed through a private 5G network, processed by FabriStack and Laminar and used to trigger OpenFOAM CFD simulations on HPC clusters at UCSB and Notre Dame.

The system demonstrated real-time data-driven simulation, adaptive resource scheduling with pilot jobs and multi-site portability—validating the vision of a digital-physical fabric that unifies sensing, wireless networking and large-scale computing for near real-time agricultural decision support.

IRI demonstration: live-streaming of raw DAQ data from FNAL to NERSC for real-time analysis, with an expected flow at ~50Gbps, to explore how IRI can address the needs of triggerless or streaming DAQ architectures.

This demonstration showcases the power and flexibility of ANARI, the Khronos standard for portable, high-performance rendering. This demonstration will show how ANARI enables scientists and developers to write visualization code once and run it efficiently on a variety of rendering backends, from CPUs to GPUs, without changing their application.

The session will feature live examples running on supercomputing resources, highlighting performance portability, scalability, and ease of integration into existing HPC workflows.

IRI demonstration: live-streaming of raw DAQ data from FNAL to NERSC for real-time analysis, with an expected flow at ~50Gbps, to explore how IRI can address the needs of triggerless or streaming DAQ architectures.

The Facility for Rare Isotope Beams (FRIB) is a DOE Office of Science user facility focused on studying problems of national interest in low-energy nuclear physics. Real-time or near real-time (nearline) analysis methods are critical tools for enabling FRIB science as new detectors and data acquisition technologies which allow for higher data rates and volumes are incorporated into laboratory systems. In addition, many experiments rely on computationally intensive analysis tasks where real-time processing on local computer systems is not feasible.

EJFAT/E2SAR provides a convenient platform for streaming data from FRIB to offsite locations, for example, to NERSC, where more computational resources are available to experimenters. An automated workflow was developed to stream data from an FRIB experiment to NERSC, process digitized detector waveform traces at NERSC, extract features of interest from the trace data and send the results back to FRIB.

This demonstration shows an example of how EJFAT/E2SAR can be used in future applications to connect the data stream from a live experiment to external high-performance computing resources.

The Flux Framework is a suite of distributed services that provide resource management and scheduling for any Linux system, from a laptop to the world’s largest supercomputers. Flux’s ability to run locally in containers provides a layer of portability for application developers and scientific programmers.

In this demo, we’ll explore Flux’s ability to run locally with mpibind’s support for affinity and binding. We’ll explain how to get Flux containers through docker hub and work with binding on just a laptop — no supercomputer required.

Presenting CelloAI, a locally hosted coding assistant that leverages Large Language Models (LLMs) with retrieval-augmented generation (RAG) to support HEP code documentation and generation. This local deployment ensures data privacy, eliminates recurring costs and provides access to large context windows without external dependencies.

CelloAI addresses two primary use cases, code documentation and code generation, through specialized components. For code documentation, the assistant provides: (a) Doxygen style comment generation for all functions and classes by retrieving relevant information from RAG sources (papers, posters, presentations), (b) file-level summary generation, and (c) an interactive chatbot for code comprehension queries. For code generation, CelloAI employs syntax-aware chunking strategies that preserve syntactic boundaries during embedding, improving retrieval accuracy in large codebases. The system integrates callgraph knowledge to maintain dependency awareness during code modifications and provides AI-generated suggestions for performance optimization and accurate refactoring.

This demonstration will evaluate CelloAI using real-world HEP applications from ATLAS, CMS, and DUNE experiments, comparing different embedding models for code retrieval effectiveness. Results demonstrate the AI assistant’s capability to enhance code understanding and support reliable code generation while maintaining the transparency and safety requirements essential for scientific computing environments.

Under the DOE-MEXT agreement researchers are building a prototype of a framework that allows scientific applications to integrate AI/ML engines in the workflow.

OLCF’s Slate platform is designed to support Kubernetes workloads and enables researchers to self-manage their open and moderate enclave deployments such as databases, message brokers, web portals, etc., through a low barrier web interface.

The talk will conclude with a demonstration of an end-to-end workflow that includes launching a database from the Slate web portal, data generation on Frontier, OLCF filesystem access from Slate’s JupyterHub platform, data ingestion into the database, and viewing that data from a user friendly front-end or via traditional queries.

Come talk to a Flux developer and have your questions answered!

The National Environmental Policy Act (NEPA) requires federal agencies to conduct environmental reviews for proposed major federal actions disclosing the potential environmental impact of actions. This process has been generating a vast amount of review documents that contain rich information and potential to inform and streamline future review process. However, the document repositories are scattered across multiple databases and organizations. Furthermore, these reviews typically require enormous amounts of human labor and time, creating significant bottlenecks that can delay project completion by months or even years. To address these challenges, the PermitAI project is developing a suite of AI web applications to streamline environmental permitting through innovative technology.

This demo will showcase SearchNEPA and CommentNEPA, the two flagship research prototype solutions from the PermitAI suite which are currently being tested by several federal users in beta mode. SearchNEPA provides advanced discovery capabilities across approximately 140,000 NEPA documents, including Environmental Impact Statements, Environmental Assessments and Categorical Exclusions. The platform features customizable filters, extensive metadata and integrated chatbot functionality that allows users to quickly retrieve information from this massive database. CommentNEPA employs a Human-AI agentic framework to autonomously process public correspondence letters, delineate comments, extract concerns and categorize them systematically into bins. The system incorporates AI and human feedback loops to continuously optimize AI prompts and improve generated summaries and concern extraction accuracy.

Both tools support the Department of Energy’s mission to convert vast data repositories into AI-ready resources and accelerate energy infrastructure deployment by streamlining permitting processes with AI technology. These applications aim to help reduce project timelines while maintaining thorough environmental oversight.

In this demo a crew of AI agents configure, start, monitor and control the 3D printing of a metal part. Using a new agent-in-the-loop control methodology, this demonstration showcases how researchers can inject human intuition into control decisions and make on-the-fly changes to control logic in response to new information. The agents use a combination of human guidance, in-situ monitoring data and simulations to make time-sensitive decisions regarding the printer parameters for each layer to control the material properties of the metal part. A human operator manages the workflow from a chat interface, seamlessly directing a workflow that bridges the cloud (AI agents), edge (3D printer) and the Oak Ridge Leadership Computing Facility (simulations).

This demonstration will operate with a digital twin of the 3D printer in the place of a real printer, but will have videos and printed parts of the system being used to print real parts at Oak Ridge National Laboratory’s Manufacturing Demonstration Facility.

The demonstration will consist of a series of interactive (simulated) prints with visuals of the chat interface, the digital twin simulation of the print with synthetic infrared camera feeds, the results of forward-looking simulations to inform decisions and the reasoning of the agents making control decisions.

Presenting the Quantum Chip Hero Run, a full‑chip, time‑domain electromagnetic simulation of an entire superconducting quantum processor performed on the full DOE Perlmutter system. Using ARTEMIS, the open‑source, massively parallel time-domain solver, researchers capture realistic control/readout pulse dynamics and system‑scale wave phenomena across the complete 3D stack, from chip‑scale routing down to micron‑scale features.

This workflow places numerical “probes” throughout the design to quantify local fields and resonances, exposing spurious modes, and coupling effects that steady‑state or sub‑system models miss. The result is a unique, predictive capability that closes the loop between classical EM simulation and quantum‑chip design — enabling co‑optimization of control lines, readout networks, filtering, and packaging and accelerating the path to higher‑fidelity, lower‑crosstalk processors.

The PanDA workload management system, central to large-scale distributed computing in high energy physics and nuclear physics, is being extended with the Model Context Protocol (MCP) to enable intelligent, tool-based artificial intelligence (AI) interactions with PanDA services. The resulting AskPanDA system, developed under the DOE-ASCR-funded REDWOOD project, functions as an AI operations agent that provides natural language access to PanDA through specialized MCP clients. These clients support job and task diagnostics, log-based failure analysis and documentation retrieval using Retrieval-Augmented Generation (RAG).

This demonstration presents the Ask PanDA architecture and shows how MCP enables large language models to interact seamlessly with distributed workflows. Live examples will highlight dynamic error triage, metadata interpretation and knowledge-base queries, illustrating how AI-driven tools can enhance operational intelligence in exascale computing environments.