Speaker
Description
An agentic framework provides a practical pathway for facility-specific AI adoption by enabling piecewise implementation of capabilities that can be incrementally deployed, evaluated, and extended as operational needs and AI technology evolve. This work outlines the current effort toward an AI agentic framework for LANSCE (Los Alamos Neutron Science Center), emphasizing the adaption of the LANSCE-specific knowledge. The framework is structured around several foundational layers: codified facility knowledge, access to real time EPICS signals, historical archiver data, diagnostic data analyses, operational logs, and simulation capabilities. These elements form a modular architecture that supports future extensibility through an incremental design philosophy. Rather than simple conversational interfaces, the framework focuses on workflow driven, multi step reasoning that mirrors how operators and system experts approach complex problems. A series of testing scenarios at varying difficulty levels demonstrates how combinations of tools of data retrieval, diagnostic analysis, simulation queries, and rule based reasoning can be orchestrated by AI agents to support operational tasks. This structured, agentic approach aims to move AI from a passive information source to an operator assistant in accelerator operations, providing a pathway toward enhanced reliability and informed decision support at LANSCE.