LLM-Driven Adjustments in Serious Games: A Feasibility Analysis

Background: Serious games (SGs) and telerehabilitation play a key role in the recovery of lost functions in neurological patients, with personalisation and difficulty adjustment being essential features. Objectives: This work investigates the feasibility of integrating a large language model (LLM) into an Assessment Serious Game (ASG) to analyse exercise data and recommend personalised rehabilitation programs. Methods: Medical knowledge was acquired through meetings with professionals to identify target pathologies and parameters. The ASG was integrated with GroqCloud; the prompt is designed to act as physiotherapist and SG developer to make real-time adjustments and suggest the setting configurations of other SGs. A preliminary test assessed the system's capabilities. Results: The LLM effectively recognises real-time adjustments and follows instructions for SGs parameter settings. However, limitations remain in the degree of adjustments and numerical parameter suggestions. Conclusion: The analysis demonstrates the feasibility of a designed LLM prompt to adjust SG difficulty and recommend setup parameters, while highlighting areas for improvement in reliability and accuracy.