Abstract Unsuitable physical environments are increasingly recognised not just as a nuisance, but also as a significant determinant in the pathophysiology of sleep disorders and chronic illness. According to the literature, even little environmental changes can have a significant impact on sleep homeostasis: persistent exposure to background noise above 30 decibels causes autonomic arousal associated to cardiovascular disease, but even low-level artificial light (5–10 lux) suppresses melatonin and disrupts metabolic homeostasis. When paired with thermal stress outside the ideal 18–21 $$^\circ $$ ∘ C, window or ventilation deficiencies enabling carbon dioxide to increase over 1,000 ppm, these environmental stressors affect sleep architecture and impede long-term cognitive recovery. This paper introduces an innovative approach to automatically regulating environmental conditions to ensure proper sleep. This approach leverages an integrated framework of model predictive control, fuzzy logic, and reinforcement learning. To validate this deterministic approach, the study utilises a high-fidelity digital twin of student accommodations at the Mărăşti Student Campus of the Technical University of Cluj-Napoca. Experimental results demonstrate that the proposed Meta-Controller significantly enhances physiological outcomes, yielding a notable 10.06% improvement in objective sleep quality metrics and a 5.41% reduction in health risk indicators associated with sleep deprivation. By achieving an optimised sleep score in 99.24% of cases, the study underscores the efficacy of merging heuristic logic with predictive and adaptive control paradigms. This work provides a pioneering contribution to the field of cyber-physical systems, laying a robust foundation for future advancements in environmental modelling and the development of intelligent, health-centric living spaces through advanced system dynamics.