Abstract: The core component of virtually any dialogue system is a dialog state tracker. Its purpose is to monitor dialog progress and provide a compact representation of its history in the form of a dialog state. Promising advances in the field of artificial neural networks suggest that using recurrent neural networks can substantially improve the capabilities of the state of the art dialog state tracking component in terms of training cost, accuracy and variety of the trackable information. I will introduce my dialog state tracker based on Long-Short Term Memory neural network, show how it is trained, and present some preliminary results. Also I will mention the main ideas of some of the techniques used in modern neural networks that I use.