A Novel Approach for AI Question Answer System/Agent in Bengali using LLM  

By : Susmita Koner , Saptarsi Goswami , Amlan Chakrabarti

Page No: 74-97

Abstract:
This paper proposes an innovative framework to deliver Artificial Intelligence (AI) education in Bengali, aligning with two key directives of India’s National Education Policy (NEP) 2020: (a) promoting education in regional languages, and (b) expanding AI literacy for all students. Despite growing interest, AI-related resources in Bengali remain scarce, and existing language models demonstrate limited translation accuracy for technical content. To address this gap, we present a dual-LLM solution—one fine-tuned for AI domain knowledge and the other for high-quality English–Bengali machine translation. The framework proposes to employ Retrieval Augmented Generation (RAG) to maintain an up-to-date knowledge base and improve response quality. We also explore architectural choices and design considerations to ensure robustness, accuracy, and educational relevance. Our approach aims to democratize AI education by making it accessible and engaging for Bengali-speaking learners, fostering inclusion and equal opportunity in technology education. The paper also has a case study where performance of LLMs has been evaluated using in context learning. The context has been provided in the form of textbooks available in Bengali.

Authors
Susita Koner Department of Computer Science, Bangabasi Morning College, University of Calcutta, Kolkata,  
Saptarsi Goswami Department of Computer Science, Bangabasi Morning College, University of Calcutta, Kolkata,
Amlan Chakrabarti A.K. Choudhury School of IT, University of Calcutta. Kolkata, India
 

DOI: DOI-https://doi.org/10.32381/IAPQRT.2026.50.01.4