Carolinian Chatbot: End-to-End Question-Answer Pipeline on the University of San Carlos Policies, Procedures, and Guidelines Via Retrieval Augmented Generation

The recent advancements in document embedding and context-based inference models have fueled the development and deployment of question-answering (QA) pipelines. This paper addresses a common challenge faced by universities: student inquiries regarding policies, procedures, and guidelines that often overwhelm staff and lead to delays. To address this, we present a novel End-to-End QA application employing the Retrieval-Augmented Generation (RAG) strategy. Our system leverages the University of San Carlos Student Manual 2023 and supplementary documents, along with the Nomic-Embed-Textv1.5 embedding model and Meta’s powerful LLaMA-3 inference model. Deployed on Streamlit Cloud, the application allows students to access initial responses to their questions. User interactions and retrieved data indicate reliable performance in both document retrieval and inference phases, particularly for queries spanning multiple documents. These results demonstrate the application’s potential to handle common student inquiries, thereby reducing workload on university staff and personnel.