Speaker
Description
Due to the diverse data acquisition modes and complex online analysis methods conducted at various beamlines of synchrotron radiation light sources, beamline users are often required to get acquainted with the interface, functionality and workflow of the data acquisition software before the experiment starts. Such process highly relies on the on-site guidance from the beamline staff themselves with routine explanation on instructional documentation to the users, which can be laborious and time-consuming. Here, based on a large language model, we propose an AI agent as an intelligent assistant system of the data acquisition software for synchrotron light source beamlines. The system utilizes large language model as a command hub to parse the beamline users’ linguistic description of the experimental process, assisting users to customize experimental process and parameters setting through its knowledge base and comprehension capabilities. The system interactively exchanges information and data with Mamba, the data acquisition software framework for fourth-generation synchrotron radiation source (HEPS), to facilitate experimental process control in an automated and intelligent manner that significantly lowers the learning curve of the data acquisition software. Additionally, the system provides an intelligent Q&A functionality based on large language model to assist beamline experiments. Through interactive Q&A sessions, the required material, structural and physical information can be obtained from the pre-trained large language model and the self-built knowledge base. Also, together with the experimental data, the system treats the queried information as metadata and deposits it into files through information exchange with Mamba, further enhancing the data’s completeness and usability.
| Abstract publication | I agree that the abstract will be published on the web site |
|---|
