This paper introduces a novel framework called Contrastive Uncertainty Domain Generalisation Network (CUDGNet) to enhance the domain generalisation capabilities of models trained on a single domain. The key idea is to augment the source capacity in both input and label spaces through a fictitious domain generator, and to jointly learn domain-invariant representations of each class using contrastive learning. Extensive experiments on two Single Source Domain Generalisation (SSDG) datasets, CIFAR-10-C and PACS, demonstrate that CUDGNet outperforms state-of-the-art single-DG methods by up to 7.08%. The method also provides efficient uncertainty estimation at inference time from a single forward pass through the generator subnetwork. The main contributions include a novel framework that leverages adversarial data augmentation and style transfer for domain expansion, while ensuring semantic information preservation through contrastive learning. The framework can estimate uncertainty in a single forward pass while achieving state-of-the-art accuracy. The performance of CUDGNet is validated through comparison and ablation studies on the SSDG datasets.This paper introduces a novel framework called Contrastive Uncertainty Domain Generalisation Network (CUDGNet) to enhance the domain generalisation capabilities of models trained on a single domain. The key idea is to augment the source capacity in both input and label spaces through a fictitious domain generator, and to jointly learn domain-invariant representations of each class using contrastive learning. Extensive experiments on two Single Source Domain Generalisation (SSDG) datasets, CIFAR-10-C and PACS, demonstrate that CUDGNet outperforms state-of-the-art single-DG methods by up to 7.08%. The method also provides efficient uncertainty estimation at inference time from a single forward pass through the generator subnetwork. The main contributions include a novel framework that leverages adversarial data augmentation and style transfer for domain expansion, while ensuring semantic information preservation through contrastive learning. The framework can estimate uncertainty in a single forward pass while achieving state-of-the-art accuracy. The performance of CUDGNet is validated through comparison and ablation studies on the SSDG datasets.