Sarcasm is an important part of communication. Sarcasm is conveyed through spoken words, facial expressions, and voice intonation. Thus, satirical remarks are disseminated daily through news headlines, social media comment sections, and tweets in our digital age. Even while sarcasm is often used in communication, it can be difficult to recognize, which frequently results in misunderstandings. Newspapers use sarcasm in their headlines a lot to draw in readers. Most of the time, readers misinterpret the news and pass along their ignorance to friends, colleagues, and other contacts because they are unable to see the irony in the headlines. Therefore, having a system that can reliably and automatically recognize sarcasm is more important than ever. Neural networks can be used to construct sarcasm detectors and investigate the process by which a computer picks up sarcastic patterns. The sequences that are classified as sarcastic or non-sarcastic provide the input for our project. These sequences are derived from a news headline dataset. We constructed an efficient model using Bi-LSTM which effectively detects sarcasm. Our model has excellent classification accuracy and can distinguish between sarcastic and non-sarcastic sentences.