Sarcasm detection in textual data remains a significant challenge in natural language processing because of contextual understanding, and capturing emotions from text is very complex. This paper proposes a context-aware model for detecting sarcasm in text, combining the power of the Universal Sentence Encoder (USE) with a Bidirectional Long Short-Term Memory (Bi-LSTM) network. The USE captures the semantic meaning coherence from sentences by embedding in to 512 dimensions, while the Bi-LSTM model can capture the sequential patterns and coherence, content for sarcasm detection from both directions. Our model achieved an accuracy of 0.91 on a benchmark sarcasm detection dataset, outperforming several traditional approaches. The results demonstrate the optimality of content level text embeddings with advanced deep learning techniques for handling complex textual nuances in sarcasm detection.