Speaker
Description
As agentic AI systems enter accelerator operations, a foundational capability is the ability to reliably translate natural-language requests into concrete control-system signals. This contribution surveys and systematizes several semantic channel-finding strategies that we have implemented and deployed across multiple accelerator facilities. We present four mature approaches—(1) in-context dictionary search, (2) hierarchical agentic navigation through middle-layer structures, (3) compositional reasoning for systematic naming schemes, and (4) knowledge-graph–based semantic search—and discuss how each addresses different scales, legacy constraints, and control system architectures. These methods form a unified strategy for building robust, facility-agnostic channel finders that can be embedded into agentic frameworks such as OSPREY. Demonstrations at UCSB FEL, the ALS, CEBAF, and XFEL-like DOOCS environments show that these approaches generalize well across laboratories and can be validated through expert-curated benchmark datasets.
| In which format do you inted to submit your paper? | LaTeX |
|---|
