Unleashing AI in Education: A Pre-Trained LLMs for Accurate and Efficient Question-Answering Systems

In today's fast-paced academic environment, efficient access to professional development resources is of paramount importance. This study investigates the integration of Retrieval-Augmented Generation (RAG) and the LangChain framework to develop a question answering (QA) system using the Llama-2 large-language model. The QA system was designed to improve information retrieval accuracy and relevance for policy-related questions based on the handbook for the faculty and non-teaching staff development program at Iloilo Science and Technology University, Philippines. By leveraging advancements in artificial intelligence and natural language processing, the system processes natural language questions and retrieves relevant information from diverse data sources such as handbook and Frequently Asked Questions (FAQs) in PDF format. The QLoRA technique was employed for model fine-tuning with 4-bit precision and the optimization of VRAM usage while retaining its high performance. The fine-tuned model indicated that the training process for question and answering reached the 105<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> global step, along with its training loss and other training metrics, including runtime, samples processed per second, steps processed per second, total floating-point operations, and epoch information. Thus, this study demonstrates that this approach enhances information accessibility and support efficiency within academic institutions, especially during times of crisis like pandemic. Future research could further enhance QA systems to augment question-answering systems by integrating personalized features, cutting-edge natural language processing techniques, affective computing techniques, machine learning strategies, and corresponding evaluation performance metrics.