<p>Sepsis is a common critical illness in intensive care medicine, affecting millions of patients globally each year. It has a high mortality rate and is one of the leading causes of death in intensive care units (ICUs). Despite significant advancements in sepsis research through artificial intelligence technologies in recent years that have reduced mortality rates, the current mortality rate in patients with sepsis remains as high as 25%. Additionally, existing prediction models suffer from a lack of transparency (the so-called “black box effect”) and fail to fully capture the complexity of perioperative data, limiting the trust that medical professionals and patients have in their predictive outcomes. This study leverages the unique capabilities of large language models to characterize perioperative sepsis monitoring data. Through representation learning, it transforms sequential patient monitoring data into textual and visual formats, thereby developing a specialized, interpretable predictive diagnostic model dedicated to sepsis treatment. This model aims to enhance the robustness and interpretability of sepsis treatment models, thereby standardizing the diagnosis and treatment processes for sepsis, reducing mortality rates, and improving patient outcomes. This study provides novel insights into the application of large language models in the medical perioperative field, offering significant scientific relevance for advancing medical artificial intelligence. It promises substantial progress in the field of critical-care medicine by enhancing the accuracy and credibility of sepsis diagnosis and treatment.</p>