Named Entity Recognition (NER), sometimes also referred to as Entity Extraction has been an integral part of Natural Language Understanding Systems. To date, there have been continuous myriad efforts in linguistics and sequential modelling which have led to the development of reliable and highly efficient NER architectures. One frequent occurrence, however, is that most work done constitutes an inclination towards resource rich corpora. To bridge the grey space for real time code-mixed expressions, we present CASCAIN (Cascaded Attention Infused Network), which leverages attention for capturing sentence level context. We also experiment with multiple architectures involving varying number of recurrent units, attention types and stacking arrangements. To provide a comprehensive performance benchmark, we train and test our models on an extended set generated from two popular Hinglish datasets. Our best performing architecture incorporates Stacked Bi-Directional LSTM Units, each of which are paired with Attention Mechanisms for generating representations from Context Vector. CASCAIN achieves an accuracy of 94.31% on the Validation Set.