Project on the paper “Neural Graph Generator: Feature-Conditioned Graph Generation Using Latent Diffusion Models”. The baseline combines a Variational Graph Autoencoder (VGAE) for latent graph encoding and a Latent Diffusion Model for generating graphs conditioned on textual descriptions. We improved this by integrating contrastive learning and enhanced encoder-decoder architectures. The task evaluates graph generation based on the Mean Absolute Error (MAE) of predicted properties.