Chatbot for Student Descipline Handbook-Related Queries: A RAG-Based LLM Using Llama-3 Approach

This study presents a Retrieval-Augmented Generation (RAG)-based chatbot designed to handle student discipline handbook-related queries using Llama-3 and ChromaDB. By employing Sentence-BERT (SBERT) for semantic similarity assessment, the chatbot ensures accurate, contextually relevant, and policy-compliant responses. Unlike conventional chatbots relying on keyword matching or rule-based systems, the proposed model integrates retrieval and generation techniques, enhancing response precision and coherence. The system is deployed using Gradio, offering a user-friendly interface for seamless interactions. Quantitative evaluation revealed a high mean similarity score of 0.9219, a median of 0.9373, and a low standard deviation of 0.0750, indicating reliable performance. A p -value of 0.0000 confirmed statistical significance, while the lowest similarity score of 0.7267 identified areas for improvement. This research demonstrates the chatbot's potential as an AIdriven educational support tool, enhancing accessibility to institutional policies and reducing administrative workload.