Synthetic data generation has emerged as a powerful technique for enhancing data-driven systems where real-world observations are sparse, privacy-sensitive, or highly imbalanced. This chapter presents a novel framework that integrates Extreme Value Analysis (EVA) with Generative Adversarial Networks (GANs) to produce synthetic data that not only captures the bulk distribution of real-world observations but also accounts for rare and seasonal extremes. The proposed methodology includes conditioning the GAN on EVA-derived temporal features such as monthly seasonality and the presence of extreme events, as well as evaluating the synthetic outputs using return level plots and generalized Pareto-based return period analysis. As a case study, we apply this method to Electric Vehicle (EV) charging data provided by DEI Blue (Greece), illustrating how synthetic charging scenarios can be realistically reproduced and evaluated. Results demonstrate that while the GAN successfully learns seasonality and general usage patterns, EVA-based evaluation reveals limitations in capturing tail-end extremes—offering a novel benchmark for synthetic data quality assessment.