Abstract
Generative Artificial Intelligence (AI) transforms financial technology (FinTech) by creating synthetic data, enhancing predictive analytics, and automating complex tasks. This paper addresses the limitations of traditional machine learning models in handling data scarcity and evolving fraud patterns in finance. We propose a novel hybrid framework that integrates Generative Adversarial Networks (GANs), Large Language Models (LLMs), and Variational Autoencoders (VAEs) to improve credit scoring, fraud detection, and financial document automation. Our method employs a Conditional Tabular GAN (CTGAN) for synthetic data generation to balance datasets, a VAE for anomaly detection in transactional data, and an LLM for generating interpretable reports and compliance documentation. Experimental results demonstrate that models trained on GAN-augmented data achieve an 8% increase in AUC for credit scoring and an 18% improvement in F1-score for fraud detection on imbalanced datasets. A dedicated compliance layer reduced demographic bias by 37%. The study confirms that a carefully designed generative AI framework can significantly enhance model performance, fairness, and operational efficiency in FinTech applications while addressing critical ethical and regulatory challenges.