Detecting emotions in Bangla text presents a considerable challenge due to factors such as code-mixing, transliteration, and the limited availability of annotated datasets. This research presents a new MuRIL-based framework that utilizes MuRIL, a transformer model pretrained on Indic languages, to tackle these issues. Our system achieves a remarkable accuracy of 92%, surpassing models like SVM + TF-IDF, BanglaBERT, and XLM-R. It performs particularly well in recognizing emotions such as anger, joy, and fear, while effectively managing code-mixed and transliterated Bangla text. This research emphasizes MuRIL's capabilities as a strong instrument for multilingual emotion classification in low-resource languages like Bangla. Notable applications of this framework encompass social media sentiment analysis, monitoring mental health, analyzing customer feedback, and real-time emotion detection in customer service and human computer interaction systems.