Spam detection in text communication has gained significant attention with the proliferation of digital communication channels. Although most prior work in this area has addressed monolingual spam filtering, there is a lack of work on Bilingual spam filtering especially for low resource languages such as Bangla. In this paper, a new approach to spam detection for both English and Bangla text messages through the application of NLP is proposed. A unique dataset of over 2500+ bilingual text messages are presented, evenly divided between spam and ham. State-of-the-art transformer models are leveraged, with XLM-RoBERTa being implemented due to its recognized multilingual capabilities, to navigate the complexities of mixed-language data. Robust preprocessing techniques, tokenization, and feature extraction are included in the methodology, followed by experiments with traditional machine learning classifiers and the fine-tuning of the pre-trained model. Comparative analysis reveals that although traditional machine learning classifiers performed reasonably well, the fine-tuned XLM-RoBERTa model demonstrated superior performance by achieving significantly higher accuracy. It effectively captured complex linguistic patterns in bilingual messages. Following extensive data preprocessing and fine-tuning steps, the model reached an accuracy of 97%, underscoring the impact of these steps on enhancing predictive capabilities. A web application offering real-time classification using a fine-tuned XLM-RoBERTa model developed for bilingual spam text detection.