The PhD project envisions creating a framework for designing and building LLM-systems tailored for human-centric, safety-critical applications. The aim is to research how to effectively integrate specialized data and domain knowledge in the decision making processes. Moreover, the research group aims to investigate how to integrate risk assessment and bias in the generated outputs. Lastly, they aim to study how to effectively integrate larger LLM systems into health care in ways guided by health care professionals for the highest on-site usefulness. The plan is to build this system as a network of LLM-agents with access to call for external tools, working to deliver safe and context-rich answers to clinical questions. Also part of the plan is to validate this framework by implementing a concrete system using real-world data for healthcare applications at the Region Västra Götaland and Sahlgrenska University Hospital. More specifically, a focus will be on the application of prognosis for heart failure patients. This entails real-world multimodal data from 78 000 hospitalizations in more than 40 000 consecutive patients.
PhD student: Erik Aerts, Chalmers