Xây dựng trợ lý ảo thông minh cho sinh viên Học viện Chính sách và Phát triển

Abstract

Students at the Academy of Policy and Development frequently require support in understanding institutional regulations and seeking advice related to their academic and personal development. In response to this need, our team developed an intelligent virtual assistant in the form of a user-friendly chatbot. Leveraging the capabilities of Large Language Models (LLMs) combined with Retrieval-Augmented Generation (RAG) techniques, the chatbot was architected as an API-based system with WebSocket integration to ensure real-time, continuous interaction. The system was implemented using Gemini 1.5 Flash for the core model, FastAPI for the backend, and a minimal React-based frontend. The chatbot has demonstrated stable performance, achieving an accuracy rate of approximately 85%. These initial results highlight the system’s potential for future development, particularly in terms of deployment flexibility and scalability to integrate additional functionalities. Building on this foundation, we are continuing to optimize the system to enhance user experience and expand its application scope.