With the rapid advancement of technology, the usage of social media and digital media has grown exponentially, driven by the widespread of the Internet. Although digital media offers countless of opportunities and serves as powerful communication means and interaction between individuals of all ages, however, it is often misused by some to inflict harm on others. This misuse contributed to the integration of harmful behaviors such as cyberbullying, hate speech, sexual harassment and other forms of online trolling. Cyberbullying is a serious issue in online environments and a growing societal concern that cannot be avoided. It poses significant risks to the mental and physical well-being of victims, leading to depression, anxiety, and suicidal thoughts. Therefore, addressing this issue requires active measures to make platforms safer and mitigate the impact of such acts. Given the complexity of online interactions, human monitoring is insufficient alone to identify instances of cyberbullying effectively. Therefore, Automatic Detection system and Mitigation strategies for Cyberbullying on social media platforms is a critical and important challenge. Technological solutions are essential in addressing such harmful trend in order to create a safer online environment for all users. This research focuses on the automatic detection of cyberbullying using a simple Bi-directional Long Short-Term Memory architecture (Bi-LSTM). Bi-LSTM networks which unlike other traditional existing methods, are capable of capturing context in both directions and enhancing the detection by effectively understanding dependencies. For example some words used alone don’t always indicate bullying behavior but when combined with other words can indicate harmful intent. Hence Twitter posts are short sentences, traditional models can miss important contextual information because they deal with word independently however on the other hand Bi-LSTM understands the relationship between words and their meaning. Natural Language Processing NLP has crucial role as an initial step to prevent such harmful behavior by processing, analyzing and understanding the text data. The model is trained using a dataset containing labeled instances of both bullying and non-bullying tweets. The performance of the system will be validated through experiments conducted on the most popular platforms such as Twitter showing its high precision in detecting the Cyberbullying. This work also investigates the preprocessing of Twitter data, feature extraction and model evaluation metrics such as accuracy, recall, precision and F1- score. The results demonstrate that Bi-LSTM surpasses traditional methods previously used for detection and contributing significantly to the development of solutions for identifying online harassment detection. Keywords: Cyberbullying Detection, Social media, Natural Language Processing, Deep Learning Model, Bi-LSTM.