Any Software Engineering projects including Machine learning solutions require adequate data. Relevant data gathering is a very tedious process. Data privacy concern has made it more difficult. Generative adversarial networks (GANs) can generate synthetic data that mimics real data distributions while preserving privacy, but they tend to lose the topological and relational features of the original data. This affects the usefulness of synthetic data for real-world applications. This work proposes to use topological data analysis (TDA) techniques to enhance GAN training and improve the quality of generated tabular synthetic data. This research introduces novel TDA-based topological loss functions, such as persistence diagram loss and homological connectivity loss, to regularize GAN training objectives. The main contributions of this paper are the development of novel topological loss formulations, adaptive weight tuning mechanisms for topological losses, controlled integration techniques, and detailed analysis of the effects on GAN learning. The models enhanced with TDA attributes consistently outperformed the base model across various Metrics. Specifically, the Homological Connectivity Loss Score, Persistence Diagram Entropy Loss Score, Persistence Diagram Loss Score, and Persistence Homology Loss Score all show improvements for the models with TDA attributes. By combining deep generative models with topological analysis, this research enables principled synthesis of realistic tabular data with complex dependency structures, advancing the state-of-the-art in privacy-preserving data modelling.