Project Summary
Project Caladrius is a generative AI-powered virtual patient system developed at Ateneo de Manila University to support medical history-taking practice in Philippine medical schools. While initial evaluations demonstrated that the text-based interface was sufficiently usable, user testing revealed a notable limitation: the virtual patients lacked personality distinctiveness. Specifically, patients across different personality types — neutral, talkative, and aggressive — appeared too similar in character, and the aggressive personality in particular was difficult to trigger. This proposal identifies personality distinctiveness as a critical gap in the current implementation and proposes a three-part solution: (1) replacing the underlying language model with a newer, more capable model; (2) implementing Retrieval-Augmented Generation (RAG) using a vector database to more efficiently surface relevant patient information while preserving the model’s instruction-following capacity; and (3) redesigning the system prompts to more precisely encode personality-specific behaviors. Together, these improvements are expected to produce more realistic, distinguishable patient characters that better simulate the variety of patients a medical student would encounter in clinical practice.
