Bespoke anatomy education without coding: a proof of concept for AI-driven simulation generation

Abstract Interactive digital resources are widely used in anatomy education, but their development typically requires programming expertise or commercial software, and pre-existing resources may not match an educator’s specific teaching requirements. Large language models (LLMs) are now capable of generating functional web code from conversational prompts, creating the possibility of educators producing their own interactive simulations without programming experience. Using LLMs to generate simulations for anatomical education, rather than as tutoring or assessment aids, remains under-investigated. This brief report is a proof-of-concept workflow in which three LLM platforms (OpenAI Codex, Google Gemini, and Anthropic Claude) were provided with reviewed anatomy content and prompted to generate interactive, browser-based simulations. Ten single-file Hypertext Markup Language (HTML) simulations were produced covering cranial nerve lesion testing, the pupillary light and corneal reflexes, extraocular muscle testing, visual field defects, spinal cord syndromes, the circle of willis, and the urinary sphincter function system. The simulations incorporate AI (Artificial Intelligence)-generated anatomical images, adjustable lesion parameters, and scenario-based clinical reasoning. Iterative refinement was necessary to achieve accurate results, as the AI repeatedly made errors, including confusion between left and right orientations, inaccurate positioning of facial features, and the occasional generation of images instead of modifying the code as instructed. These challenges required multiple rounds of correction and careful human oversight to ensure the final output met the intended requirements. Preliminary simulation assessment and comments from Medicine Year 2 colleagues, Oxford AI Build club, learning technologist, Teaching demonstrators were encouraging regarding clarity and engagement, although no formal student evaluation of learning outcomes has yet been conducted. We conclude that LLMs can translate structured anatomical knowledge into functional interactive anatomy simulations and describe a replicable workflow, while emphasising that expert review remains indispensable.