This document proposes using deep learning approaches to detect malicious domain names at scale. It presents an architecture that takes raw domain names as input and uses deep neural networks to obtain optimal feature representations and classify domains as benign or malicious. The approach is evaluated on two datasets, one collected internally and one with externally sourced benign and malicious domains. Results show deep learning outperformed classical machine learning algorithms, with over 97% accuracy on the internal test set. Future work could involve collecting and analyzing additional log sources using the proposed architecture to better detect malicious activities within organizations.