22nd AIAI 2026, 16 - 19 July 2026, Chania, Crete, Greece

Interactive Retrieval-Governed LLM Tutoring for Surgical Site Infection Prevention Education

Stylianou Frixos, Vakanas Konstantinos, Papageorgiou Panagiotis, Charalambous Andreas, Nicolaidou Iolie, Michael Koralia, Loizou Christos, Kyriacou Efthyvoulos, Tsitsi Theologia

Abstract:

  Surgical Site Infections (SSIs) remain a major source of postoperative morbidity, prolonged hospitalization, and increased healthcare cost [22,6]. Although Large Language Models (LLMs) offer promising capabilities for medical, patient, and nursing education [15,5,4,12], their use in clinical learning requires strict governance, traceability, and evidencebased grounding. This paper presents SurgiNurse AI, a governance-aware retrieval-augmented tutoring platform for SSI prevention education. The system integrates administrator-controlled source eligibility and verification, verified-only retrieval, Role-Based Access Control (RBAC), serverside governance enforcement, and structured generation of explanations, flashcards, and Multiple-Choice Questions (MCQs). In a preliminary expert pilot using ten representative SSI queries, the system achieved an answer correctness score of 0.90, perceived helpfulness of 4.7/5, Precision@ 1 of 0.80, Precision@3 of 0.70, Precision@6 of 0.55, and full MCQ structural compliance. These results are promising but preliminary, as they are based on limited expert review and a curated corpus whose coverage may influence system performance. Observed failure cases mainly involved reduced retrieval relevance in larger candidate sets and minor inaccuracies when the corpus did not fully cover the user request.  

*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.