This paper presents a virtual patient (VP) platform for medical education, combining a social robot, Furhat, with large language models (LLMs). Aimed at enhancing clinical reasoning (CR) training, particularly in rheumatology, this approach introduces more interactive and realistic patient simulations. The use of LLMs both for driving the dialogue, but also for the expression of emotions in the robot's face, as well as automatic analysis and generation of feedback to the student, is discussed. The platform's effectiveness was tested in a pilot study with 15 medical students, comparing it against a traditional semi-linear VP platform. The evaluation indicates a preference for the robot platform in terms of authenticity and learning effect. We conclude that this novel integration of a social robot and LLMs in VP simulations shows potential in medical education, offering a more engaging learning experience.