A Retrieval-Augmented Multilingual Chatbot for Intelligent College Information Services
TL;DR
The experimental results obtained were 92% response accuracy, 90% retrieval relevance, 1.8 s response time on average, and 91% user satisfaction, thus proving the efficiency of the suggested architecture in providing reliable and scalable academic information services.
Abstract
The study proposes an artificial intelligence-based multilingual university chatbot by utilizing the Retrieval-Augmented Generation (RAG) approach for intelligent institutional information retrieval. The developed chatbot leverages Sentence Transformer embeddings, FAISS-enabled semantic search, and the Llama-3 large language model for generating context-specific answers that depend on the institutional knowledge base. To ensure high availability, the chatbot offers support for multiple languages by enabling language translation and voice communication based on the Whisper speech-to-text model. Other functionalities of the chatbot include user authentication, image and document search, chat history management, and campus navigation. The proposed system was developed for Rajeev Gandhi Memorial College of Engineering and Technology (RGMCET), and the experiment was carried out with representative queries from the institution. The experimental results obtained were 92% response accuracy, 90% retrieval relevance, 1.8 s response time on average, and 91% user satisfaction, thus proving the efficiency of the suggested architecture in providing reliable and scalable academic information services.