I work in a a Belgian consulting company specialised in AI, which has a partnership with Google. Therefore, we use Google Cloud Platform (GCP) infrastructures and services for nearly all of our projects. I have done some research and model developments for automatic expense validation. Enable automatic prediction of expense category and expense total amount, in order to facilitate expense validation. I have compared two types of Deep Learning models: text-based and image-based models. In both cases, I performed transfer-learning on top of pre-trained models (BERT, for NLP, and Inception V3, for image processing), using Tensorflow. In parallel I decided to focus only on expense categories with low intra-class variance, in order to achieve decent prediction accuracy.