<![CDATA[is study aims to develop an automotive chatbot system capable of providing contextual, accurate, and reference-based responses using the Retrieval-augmented generation (RAG) approach. Conventional automotive information services often face limitations in delivering relevant technical information quickly and accurately. Large Language Models (LLMs) provide natural responses but still suffer from hallucination issues that may reduce answer reliability.To address this limitation, the proposed system integrates semantic retrieval using Multilingual E5 and reranking using DeBERTa v3 before generating responses through a Large Language Model. User queries and automotive documents are transformed into semantic vector representations, enabling retrieval based on contextual similarity rather than keyword matching. Retrieved documents are subsequently reranked to improve contextual relevance before being used as supporting knowledge during answer Generation.The system is implemented as a web-based application and evaluated using retrieval metrics and the RAGAS framework. Experimental results indicate that the proposed approach improves contextual understanding and answer relevance while reducing the possibility of hallucinated responses. The developed chatbot demonstrates the potential of Retrieval-augmented generation for supporting intelligent automotive information services. The evaluation results show that the proposed system achieved a Faithfulness score of 1.000, an Answer Relevancy score of 0.883, and a Context Precision score of 1.000. These results indicate that the integration of Multilingual E5 and DeBERTa v3 successfully improves contextual relevance while reducing the risk of hallucinated responses]]>
Copyrights © 2026
