Adaptive and personalized AI systems in healthcare rely on user and situational data to provide optimal support. However, patients facing illness, pain, or complex therapeutic decisions often experience cognitive impairments, leading to incomplete or biased data that hinder informed decision-making. The FEDWELL project explores artificial mental models (AMMs) in adaptive AI systems to address these challenges, focusing on rehabilitation after knee/hip surgeries and therapy decisions for cognitively impaired patients (e.g., MS, dementia). By integrating AMMs, structured surveys, contextual data, and machine learning, FEDWELL aims to align AI recommendations with patient needs and preferences.