A reversible transformer based Bangla conversational agent

Abstract

Chatbots are becoming increasingly popular on various industrial platforms as AI progresses and the demand for an online conversation system rises. Transformer model-based intelligent tools pay close attention to their applications since they may mimic human behaviour in plain language. Transformer models confront several difficulties when handling lengthy sequences, thus we developed a novel approach by employing reversible transformer models. In this paper, we tried to develop a model for the Bangla chatbot based on the Transformer algorithm and Trax framework. Based on the reversible transformer model, the suggested framework's performance on MultiWoz 2.1 and our unique dataset সমাচার (Shomachar) validates its superiority with a 98% accuracy rate. © 2026 selection and editorial matter, Nazmul Siddique, Mohammad Shamsul Arefin, K. M. Azharul Hasan, and M. Shamim Kaiser; individual chapters, the contributors