Applying RAG to University Curriculum Book

Abstract

Generative AI is a powerful tool with various emerging applications. However, challenges such as hallucinations and lack of domain-specific knowledge remain to be addressed. Retrieval-Augmented Generation (RAG) has been developed as one solution to these challenges. This study focuses on applying RAG to text information. The approach involves splitting text into relatively short chunks, vectorizing them, and enabling fast vector searches. This research reports experimental findings on the impact of chunk length on final accuracy and the influence of ambiguity in the original text.