UNI ASSIST-AI: An Intelligent University Assistant Chatbot with LLM, RAG, and Multimodal Capabilities

Key Points

• The aim is to develop an intelligent university assistant that accurately responds to student queries using verified institutional information.
• Developed a chatbot utilizing Retrieval-Augmented Generation (RAG) methodology.
• Integrated a semantic vector retrieval pipeline with a GPT-based generative model.
• Enabled multimodal input capabilities for text, voice, and image queries.
• Evaluated the system using precision, recall, and F1-scores against rule-based and vanilla LLM systems.
• Achieved a Precision of 0.87, indicating high accuracy in relevant responses.
• Obtained a Recall of 0.84, demonstrating good ability to identify relevant information.
• Reached an F1-score of 0.85, reflecting a balance between precision and recall.

Abstract