PROJECT SUMMARY Sepsis kills an estimated 11 million people worldwide every year. Sepsis also contributes to as many as 50% of US hospital deaths. Early treatment is the only universally recognized modifiable factor for improving sepsis mortality. Thus, current treatment guidelines focus heavily on early sepsis care immediately following hospital presentation. However, today, little is known about even earlier, pre-hospital, opportunities to predict, recognize, or treat sepsis. Our research program has deployed innovative translational informatics approaches to elucidate the presentation, pace, and profile of infection in pre-sepsis patients that could enable novel pre- hospital approaches designed to mitigate, or even prevent, sepsis. These pre-sepsis opportunities are now being recognized as a key new frontier of sepsis research and care. In this renewal, we are extending successful research that has begun to identify and characterize pre-sepsis opportunities for early, targeted intervention. To achieve this goal, we have already leveraged granular and extensive electronic health record (EHR) data and machine learning methods. This supplement seeks to enhance our computing capabilities by leveraging powerful GPU architecture and high-speed storage/networking to now incorporate imaging, text, and time series data into our predictive models through applying open-source deep learning and large language model techniques. This will augment our capability to develop pre-sepsis predictive models that are efficient, effective, and equitable. Our findings will have broad and immediate implications for patients, clinicians, and health systems to address this devastating condition. These results will also inform the design of sepsis public health programs and future prospective interventional studies aimed at improving outcomes for a condition that remains common, deadly, and costly.