It has always been our dream to create machines that can think. In the early phase of artificial intelligence, we solved the problems that could have been described by a list of formal and mathematical rules. The main motivation was solving the tasks that are trivial for people to perform but difficult to describe formally, for example recognizing objects in images or spoken words. Neural networks have been one predominant area of research in artificial intelligence that stayed in limelight over the past five decades or so. However, one major boost in this domain happened around 2010 when the computers got much faster and data sets got much bigger. The existing algorithms with some critical tweaks and improvements became much more efficient and won several challenges in computer vision, speech recognition, and natural language processing. In this talk, I would briefly describe how neural networks works and then move towards state-of-the-art approaches of deep learning. I would describe in detail how to build deep learning models and its training procedures using the deep learning library "Keras" that removes all of the complexity, leaving you an API containing only what you need to know to efficiently develop and evaluate neural network models. This talk aims to present two use cases (facial recognition and sequence prediction) that uses Keras for model building and evaluation.