A RAG-Based Automotive Sector AI Assistant for Enhanced Information Retrieval

Advanced Artificial Intelligence (AI) technologies are increasingly used to ensure fast and accurate information access across industries. In the automotive sector, efficient querying of technical data by customers and service staff is essential. Due to the volume of manuals, maintenance logs, and troubleshooting guides, manual search is often impractical. This study introduces an AI-based assistant for the automotive domain, built on a pretrained language model using the Retrieval-Augmented Generation (RAG) framework. RAG improves text generation by retrieving relevant data from external sources—such as document repositories and databases— rather than relying solely on a generative model. This hybrid approach reduces hallucinations and increases response accuracy. Unlike traditional chatbots, our system draws domain-specific content from curated technical documents, ensuring higher relevance and reliability. The assistant is not a new model but a domain-specific application that integrates an existing LLM with the RAG framework for an industrial use case. The automotive assistant is designed to extract information from technical documents to deliver accurate answers to common user problems. It supports both vehicle owners and service professionals by providing real-time, context-aware information for troubleshooting and maintenance. To evaluate its performance, a validation dataset comprising 487 real customer service call transcripts (2,578 sentences, 6,445 seconds) was used. These transcripts served solely for evaluation purposes, testing the assistant's ability to generate contextually appropriate responses to realworld queries. This study demonstrates how a RAG-based model can be optimized for domain-specific use, improving information retrieval in the automotive sector. By combining retrieval and generation, the assistant enhances the accuracy and efficiency of technical support. The system was first piloted internally by call center staff, allowing for a thorough evaluation of its accuracy, safety, and compliance with responsible AI principles. Pilot results showed that the assistant significantly enhanced the efficiency and accuracy of information retrieval in technical support, improving operational performance and user satisfaction. Evaluations confirmed that it provides more precise and context-aware responses than traditional generative models, leading to a better user experience. As a result, the assistant serves as a valuable tool for both end-users and service teams, reducing time spent on searching critical maintenance information and boosting customer satisfaction.