Artificial Intelligence (AI) is increasingly applied in healthcare, but projects often encounter formidable technical and practical challenges. This article examines these challenges through the lens of three international scientific projects – CareWare, Inno4health and REMO – which implemented or will implement AI for wearable systems, health monitoring and rehabilitation. Key technical obstacles are identified in AI modelling as data quality, sensor integration, interoperability, and system validation. AI models in healthcare must contend with limited and heterogeneous datasets, high variability in physiological signals, and the need for interpretability and personalization. Data quality issues – such as sensor noise, missing data, and inconsistent data collection – can undermine model reliability. Integrating multiple sensors and ensuring interoperability between devices and health information systems remain difficult due to a lack of common standards. Validation of AI-driven systems is also challenging, requiring rigorous clinical evaluation and compliance with strict regulations to ensure safety and effectiveness. Specific technical barriers observed in project development (e.g., noisy wearable device signals, variability in vital signs, limitations of cognitive feedback systems and real-time analytics for remote care) are highlighted with examples from the projects. The article concludes recommendations for future development and research, emphasizing improvements in data quality standards, better sensor and system interoperability, robust validation frameworks in healthcare.