Retrieval-Augmented Generation Chatbot for Academic Information Services Using GROQ API

Abstract

Academic information services in universities are still largely carried out manually by administrative staff with limited working hours, making it difficult for students to receive quick responses, especially outside office hours. This study aims to design and implement an academic chatbot using the Retrieval-Augmented Generation (RAG) approach through the Groq API and a Streamlit interface. The system was developed following the Design Science Research (DSR) methodology, which includes problem identification, defining solution objectives, dataset collection, artifact design and development, system demonstration, testing and evaluation, as well as communication of research results. Black-box testing on 20 questions yielded 18 valid answers, while semantic quality evaluation using BERTScore produced a precision of 0.7054, recall of 0.7176, and F1-score of 0.7095. The average system response time was 1.45 seconds, with a median of 0.85 seconds, indicates that the majority of responses were provided in less than one second, despite a single outlier of 12.13 seconds. These results finding suggest that the integration of RAG with the Groq API holds potential for enhancing digital academic information services in higher education. Nevertheless, its operational benefits in real-world scenarios warrant further evaluation through empirical user testing.