<![CDATA[The utilization of Large Language Models (LLMs) in higher education offers significant efficiency, yet it introduces critical risks of information hallucination and conceptual bias, particularly in the discipline of Sociology. This study aims to evaluate the performance of a Retrieval-Augmented Generation (RAG) system based on the low-code platform n8n as a robust solution for hallucina-tion mitigation. The system integrates semantic search using Supabase as a vector database and the Gemini 2.5 Flash model to restrict response generation exclusively to verified academic litera-ture. The research employed an Experimental Single-System Evaluation method with a du-al-evaluation approach (quantitative and qualitative) across 50 test instruments. Quantitative testing using the ROUGE-L metric recorded a mean score of 0.354, indicating adequate structural similarity despite variations inherent to the paraphrasing nature of LLMs in analytical tasks. Thematic qualitative analysis of evaluator comments revealed 98.2% positive sentiment, with the dominant theme being “Conceptually Accurate”. Ultimately, 100% of the expert panel declared the system suitable for implementation as a reliable supplementary learning medium.]]>
Copyrights © 2026
