The paper proposes a new direction in sentiment analysis of the Bang-lish text by combining the strengths of BERT and CNN to handle the intricacies inherent in code-mixed languages. The aim was to develop a robust model that can capture sentiments in Bang-lish with high precision, and it does so with flying colors. The proposed hybrid BERT-CNN model achieved an accuracy of 94.5% in binary sentiment classification and 91.5% in multi-class classification, outperforming the traditional and standalone models with a margin, showing that the model captured the global context and local patterns in the mixed-language texts, thus very effective for sentiment analysis. All the categories of sentiments were tested extensively, and in that way, its versatility was established over other models, proving its adaptability with the subtlety that Bang-lish brings in. This piece of research will add to the field of natural language processing, hence giving a strong and state-of-the-art solution for sentiment analysis on code-mixed languages, which can help in social media monitoring, customer feedback analysis, and tracking of public opinion.