2026-07-16 –, Memorial Hall
The Atmospheric Radiation Measurement (ARM) User Facility Data Center (ADC) capable of supporting scalable, secure, and reproducible engagement with atmospheric research data is evolving towards AI-ready ecosystem. We will discuss architectural designs utilized in production scientific data setting including open-source technologies to further multi-agent coordination, agentic retrieval-augmented generation (A-RAG), shared contextual memory via vector stores, and model-agnostic inference orchestration within Kubernetes infrastructure. We will go over ARM's foundational stack designed to support agentic AI workflows for data discovery, metadata research, reasoning, and user engagement. Additionally, we will go over architectural decisions, trade-offs, and security measures pertinent to research computing environments with some demonstrations.
With large language models (LLMs) and agentic AI system becoming more prevalent, scientific data centers are investigating how this can improve data discovery, metadata interpretation/automation, and overall data-researcher interaction. Deploying LLMs alone aren't enough for advancing AI-enabled capabilities in scientific environments. We must think of a cohesive architecture that integrates with our existing research infrastructure and facilitates interoperability, reproducibility, scalability, and governance.
This talk describes the design and implementation of an agentic AI infrastructure developed within the Atmospheric Radiation Measurement (ARM) User Facility Data Center (ADC) to support AI-enabled workflows across atmospheric science data systems. Rather than developing a single application, the effort provides a foundational stack that standardizes how AI agents engage with data, tools, and users throughout the ARM ecosystem.
The architecture is organized as a layered system that facilitates modular and interoperable AI services. At its foundation is a centralized inference infrastructure providing model-agnostic access to LLMs deployed on GPU-enabled research systems. The framework introduces an Agentic Retrieval-Augmented Generation (A-RAG) approach tailored for scientific data workflows. Traditionally retrieval-augmented generation improves the accuracy of language models by grounding responses in externally retrieved information. With A-RAG, each specialized agent can retrieve domain-relevant information from ARM data services, metadata catalogs, documentation, and web services, enabling evidence-driven responses that reflect the structure and context of atmospheric research data.
The framework adopts emerging protocols such as Model Context Protocol (MCP) for structured tool access, Agent-to-Agent (A2A) for coordinated communication among agents, and Agent–User Interaction (AG-UI) protocol that support traceable conversational workflows. These protocols allow conversational interfaces, tools and applications to integrate with the framework while reusing shared services. At the central of these capabilities is shared contextual memory layer implemented through persistent vector stores that hold embeddings of structured scientific artifacts and documentation. Through this contextual layer the agents can operate over a consistent state which in turn supports coherent reasoning across sessions and workflows.
Attendees will learn about architectural patterns for building and developing agentic AI infrastructure, strategies for extending traditional RAG into coordinated multi-agent systems, and practical considerations for deploying open-source LLM tooling in environments that require security, governance, and reproducibility.
Intended audience: Software Engineers, Architects, Maintainers or Practitioners interested in AI and enabling that in scientific platforms.
While the implementation is grounded towards atmospheric science domain, the architectural principles presented are broadly applicable to other scientific data repositories, national laboratory computing environments, university research platforms, and open-source projects that aim to create interoperable and trustworthy AI-enabled workflows. Towards the end of presentation will have a demonstration illustrating how these architectural components enable coordinated AI agents to facilitate scientific data exploration in a production setting such ADC.
Chirag Shah is an Environmental Data Science Engineer and full-stack software developer working with the U.S. Department of Energy’s Atmospheric Radiation Measurement (ARM) User Facility Data Center. His work focuses on building scalable scientific data systems that improve the discoverability, accessibility, and usability of large-scale atmospheric and environmental observations.
At the ARM Data Center, Chirag leads the design and development of modern research software platforms used by scientists to explore, analyze, and interact with complex observational datasets.
Chirag's technical interests span scientific data management, distributed systems, artificial intelligence, machine learning, and advanced data visualization. His work emphasizes building robust infrastructure and user-centric tools that enable researchers to efficiently work with large observational datasets and accelerate scientific discovery in Earth and environmental systems research.
Committed to advancing modern research software practices, Chirag actively explores emerging technologies that enhance the way scientists interact with complex data ecosystems.
Utkarsh Mahai is a full-stack software engineer at the Department of Energy's Atmospheric Radiation Measurement (ARM) User Facility Data Center. He works on building software, tools, and applications that help scientists and researchers access data, streamline workflows, and focus more on advancing their science.
His work spans the full software development lifecycle, from designing user experiences and building web applications to developing backend services and integrating emerging technologies where they can provide meaningful value. More recently, he has been involved in building agentic systems, modernizing user interfaces in the age of AI, and improving the ways information and context flow through applications.
Outside of work, Utkarsh is interested in conversations around AI ethics, governance, and the broader impact of emerging technologies. He enjoys continuous learning and exploring new ideas, tools, and approaches to solving problems.
Before joining the ARM Data Center, he worked on software products and business processes in the financial technology (fintech) and entertainment industries.
Austin Aguilar is a Software Engineer for the Atmospheric Radiation Measurement (ARM) Data Center at Oak Ridge National Laboratory. Austin helps develop and maintains a variety of user facing applications that allow researchers to focus on their science rather than data discovery. Austin is also very passionate about researching and developing agentic AI workflows to be leveraged within the scope of science.