Hate speech detection is a vital task in the context of content moderation, aiming to identify and mitigate harmful language in online platforms. This abstract presents a methodology utilizing Recurrent Neural Networks (RNNs) for hate speech detection. We have processed a heterogeneous dataset containing occurrences of hate speech and non-hate speech & preprocessed the text data, and represent words as vectors using pre-trained word embedding’s. Our chosen RNN architecture, adept at capturing sequential dependencies, processes the contextual information within the text. During training, the model optimizes using a binary cross-entropy loss function and undergoes validation for hyper parameter tuning. The effectiveness of the RNN-based hate speech detection model is evaluated using accuracy, F1 score, precision and recall. This methodology provides a robust framework for combating hate speech in digital spaces, contributing to a safer and more inclusive online. We have compared our work with various base line methods with several parameters and achieved an accuracy of 96.89%.