How to Build an AI Tutor that Can Adapt to Any Course and Provide Accurate Answers Using Large Language Model and Retrieval-Augmented Generation (2311.17696v4)

Abstract: This paper proposes a low-code solution to build an AI tutor that leverages advanced AI techniques to provide accurate and contextually relevant responses in a personalized learning environment. The OpenAI Assistants API allows AI Tutor to easily embed, store, retrieve, and manage files and chat history, enabling a low-code solution. LLMs and Retrieval-Augmented Generation (RAG) technology generate sophisticated answers based on course-specific materials. The application efficiently organizes and retrieves relevant information through vector embedding and similarity-based retrieval algorithms. The AI Tutor prototype demonstrates its ability to generate relevant, accurate answers with source citations. It represents a significant advancement in technology-enhanced tutoring systems, democratizing access to high-quality, customized educational support in higher education.