RasPatient Pi: A Low-Cost Customizable LLM-Based Virtual Standardized Patient Simulator

Simulation is an essential part in medical and nursing education, allowing students to develop and be assessed on specific skills within a safe environment. Simulated or standardized patients are useful to practice communication skills, but the availability and medical knowledge of these actors can become a bottleneck. Virtual patients, on the other hand, often in the form of serious games, allow to check clinical reasoning skills. However, they oftentimes do not enable teachers to create or customize scenarios. The linguistic power of Large Language Models (LLMs) and their clinical potential allow for meaningful dialogues with a healthcare student for simulation purposes. In this paper, we present RasPatient Pi, a low-cost customizable LLM-based virtual standardized patient simulator. The simulator leverages automatic speech recognition, LLMs and text-to-speech engines. Scenarios can be specified by the teacher through a short description, while the clinical knowledge of LLMs is used to coherently complete any gap in the scenario. It can be deployed on a single-board computer to be used alongside a manikin or played in a browser, relying on a 3D avatar.