Large Language Models for Intensity Frontier Experiment Documentation and Operations

Large language models are rapidly advancing, potentially capable of functioning as specialized assistants for scientific research. We are exploring the use of state of the art large language models for intensity frontier documentation and operations. Specifically, using prompt engineering techniques as well as Retrieval Augmented Generation (RAG) to send relevant embedded documentation to the LLM to generate the desired response. These techniques include optimizing the format and amount of content provided to the model. Responses are assessed based on a variety of metrics including statistical and model-based evaluation frameworks. In terms of operations, providing these models with well defined functions enables their performance of basic experiment procedures. By integrating these enhanced models with APIs, we are working towards creating an interactive model, capable of assisting in monitoring a high energy physics experiment.