Optimizing Conversational Assistance for Universities: A hybrid approach with LLMs and RAG

The thesis presents an AI-based chatbot designed to help students at Ca’ Foscari University of Venice by giving accurate and precise answers to inquiries related to university topics such as regulations, processes and deadlines. The AI-Assistant is based on a large language model (LLM) that is focusing on Retrieval-Augmented Generation (RAG) to achieve a better understanding of student questions. The RAG is extracting the most relevant parts to the question asked from university documents stored locally such as policy documents and Frequently asked questions. The goal is to provide the LLM with the most accurate context in order to give correct responses, avoiding the risk of inaccuracies that can be caused by several AI tools such as ChatGPT which has limitations related to hallucinations when searching the web. The RAG-based chatbot was outperforming when tested against ChatGPT on 50 questions related to university information. Students expressed that the responses generated by this system are more correct, relevant and clear.