The exponential growth of social media platforms has resulted in a significant surge in user-generated content across many platforms. The increase in content offers a wide range of viewpoints and emotions, which strengthens the importance of sentiment analysis. Nevertheless, the absence of uniform metadata complicates the process of sentiment analysis, hence posing a challenge in effectively interpreting emotional responses from text samples. Sentiment analysis often incorporates polarity analysis to evaluate the attitude of a text, classifying it into positive, negative, or neutral attitudes. This study presents a resilient framework for sentiment analysis using two datasets: one in Hindi and one in English. The system utilizes the Bidirectional Encoder Representations from Transformers (BERT) method, renowned for its exceptional precision in sentiment classification applications. The methodology entails training BERT using the designated datasets and evaluating its performance in comparison to established methodologies. When assessing the suggested framework, the BERT model exhibited exceptional performance, with accuracy rates of 90% and 86% for the Hindi and English datasets, respectively. In contrast to other algorithms, which often exhibit poorer accuracy, the comparison highlights the efficacy of BERT in doing sentiment analysis across several languages, affirming its dependability and effectiveness in collecting and evaluating feelings from different text sources.