Large Language Model based Educational Virtual Assistant using RAG Framework

In recent times significant advancement have been made in text based chatbots or virtual assistants. This research study presents a latest and transformational approach to provide information support in university through the development of educational virtual assistant based on large language models and retrieval augmented generation framework. Moving forward, the technological advancement have increased the complexity of university systems, so finding innovative solution for students to access information on various topics such as admission processes, course selection, and campus facilities etc. is important. The proposed virtual assistant leverages transformer architecture based large language models specifically Meta-llama/Llama-2-7b-chat-hf and Mistralai/Mistral-7B-Instruct-v0.2 to generate human like text for student inquires. The system design includes a data retrieval process from university websites followed by data pre-processing. Building on retrieval augmented generation framework, the virtual assistant retrieves the most relevant data from a comprehensive university knowledge base, ensuring responses stay updated and precise. The virtual assistant was tested on a range of varieties of university related queries, and its response were evaluated using bilingual evaluation understudy score metrics. Experimental results suggest that the retrieval augmented generation-based Llama-2-7b-chat-hf provides a viable solution for addressing the challenge of providing university related information to students. The findings indicate that retrieval augmented generation based large language models hold significant potential for automating administrative support in educational institutions.