Electrocardiography (ECG) is central to cardiovascular care, and recent advances in artificial intelligence (AI) show that large pretrained ECG models can extract clinically meaningful information beyond visual interpretation and be reused across tasks such as risk stratification and disease prediction. The goal of this project is to develop large, reusable pretrained ECG models that are robust, interpretable, and suitable for clinical deployment. The project pursues four integrated objectives: (I) to develop large-scale pretraining strategies that produce transferable ECG representations; (II) to quantify their added predictive value beyond established clinical risk factors while incorporating causal considerations to address confounding, selection bias, and domain shift; (III) to ensure robustness, generalization, and interpretability across heterogeneous ECG formats, devices, and acquisition settings using multiple explainable AI techniques; and (IV) to evaluate federated learning approaches that enable ECG pretraining across institutions and cohorts. Together, these efforts establish a principled foundation for large pretrained ECG models and provide interdisciplinary training for PhD and Master’s students in machine learning and cardiovascular medicine.