An Agentic AI Framework for Training General Practitioner Student Skills

1 An Agentic AI Framework for Training General Practitioner Student Skills Victor De Marez, Jens Van Nooten, Luna De Bruyne, and Walter Daelemans This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. Abstract—Advancements in large language models offer strong potential for enhancing virtual simulated patients (VSPs) in medical education by providing scalable alternatives to resource- intensive traditional methods. However, current VSPs often strug- gle with medical accuracy, consistent roleplaying, scenario gen- eration for VSP use, and educationally structured feedback. We introduce an agentic framework for training general practitioner student skills that unifies (i) configurable, evidence-based vignette generation, (ii) controlled persona-driven patient dialogue with optional retrieval grounding, and (iii) standards-based assessment and feedback for both communication and clinical reasoning. We instantiate the framework in an interactive spoken consul- tation setting and evaluate it with medical students (N=14). Participants reported realistic and vignette-faithful dialogue, appropriate difficulty calibration, a stable personality signal, and highly useful example-rich feedback, alongside excellent overall usability. These results support agentic separation of scenario control, interaction control, and standards-based assessment as a practical pattern for building dependable and pedagogically valuable VSP training tools. Index Terms—Agentic AI, evidence-based medicine, large lan- guage models, medical education, virtual simulated patients. I. INTRODUCTION I N medical education worldwide, simulated patients (SP), which are trained actors who portray patients with prede- fined symptoms and behaviors [1], are used to teach essential skills such as history taking and communication skills, and explaining a diagnosis. They are also a traditional part of the Objective structured clinical exams (OSCE) assessment of students, in which they are meant to focus on a specific skill, so that SPs are used to systematically measure clinical and communication skills in a standardized way [2]. However, training and hiring these qualified SPs is a pro- cedure that requires substantial investments in resources and time [3]. Furthermore, despite rigorous training efforts, perfect replicability of the scenario is impossible due to inherent hu- man variance and errors. Additionally, the educational setting with an SP can be distracting and stressful due to the presence of the tutor and other students [3]. Virtual Simulated Patients (VSPs) are computer simulations of real-life patients programmed with clinical vignettes (clin- ical scenarios that include patient information) that allow a Manuscript received December 20, 2025. This research received funding from the Flemish Government under the “Onderzoeksprogramma Artifici¨ele Intelligentie (AI) Vlaanderen” programme. (Corresponding author: Victor De Marez.) Victor De Marez, Jens Van Nooten, Luna De Bruyne and Walter Daelemans are with the Center for Computational Linguistics, Psycholinguistics and Sociolinguistics (CLiPS), University of Antwerp, Antwerp, Belgium (e-mail: firstname.lastname@uantwerpen.be). This article has supplementary downloadable material available at https://doi.org/... Dashboard Neuroticism Openness Agreeableness ... Conversation history RAG EBM vector DB EBM disease information Vignette Generator Agent Patient vignette Patient vignette VSP Agent 1. Multi-step pre-process 2. Generate answer 3. Checklist post-process Student doctor Critic Agent During conversation quick, short, feedback on communication After conversation detail, long, feedback on communic. & diagnostics EBM disease information Feedback framework 1 2 3 4 Conv. history Fig. 1. Core schema of the four main contributions of our framework: clinical vignette generation, a three-step VSP generation method, personality customization, and automated feedback generation . learner to obtain a medical history, make a diagnosis, and prescribe a treatment plan [4]. Initially, VSPs were costly, and limited in realism, natural language capabilities, effectiveness and applicability. However, advances in artificial intelligence have accelerated development of VSPs [5], [6], thereby of- fering effective solutions to the aforementioned problems of SPs. Despite their initial limitations, VSPs offer multiple advantages over SPs. For instance, virtual patients can be used by unlimited learners at virtually no incremental cost, therefore being more cost-friendly and less resource-intensive [7]. This scalability allows for interaction beyond stressful educational settings, for instance from home. Moreover, VSPs can be configured to follow a predefined vignette consistently and to incorporate a larger number of case details than is typically feasible for human actors. Early VSPs were mainly comprised o

This content is AI-processed based on open access ArXiv data.