The goal of the project is to empirically measure if the ViT (Vision Transformer) model which is not pre-trained out performs in comparison against the CNN (Convolutional Neural Networks) model which is also not pre-trained, when attempting the validation contents of a new document to be classified. The image datasets (between 10 to 15 million records) being used would be generated through a pipeline and the split between the training & test would be 85 : 15.