In industrial contexts, the availability of large-scale, reliable time-series data is crucial for data-driven systems, yet often hampered by privacy constraints and affected by data scarcity and quality issues, including noise and missing values. In many real-world application domains, data collection can be costly and subject to proprietary or access restrictions. In this scenario, synthetic data generation has gained increasing attention as an effective strategy to overcome such constraints. Building on these motivations, this thesis focuses on the design and implementation of an end-to-end pipeline for generating synthetic data from irregularly sampled time-series associated with electric vehicle battery charging and discharging events, with the aim of producing high-fidelity and representative synthetic samples that preserve the temporal structure and statistical properties of real-world data.