A Framework for Multi-Agent LLMs in the Automotive Domain

This thesis presents the design and implementation of a large language model (LLM)–based system for the automotive domain, aimed at improving access to proprietary knowledge. Companies in this sector may struggle with large and complex datasets that are typically accessible only to technical experts. The proposed system enables non-technical users to retrieve relevant information through natural language queries, simplifying access to complex data sources. The architecture is built around a hierarchical network of LLM agents, each dedicated to a specific function. In the current implementation, a Vehicle Basket Agent and a User Preferences Agent are developed to demonstrate this approach. Particular emphasis is placed on modularity, ensuring that the system can be extended with additional agents as new requirements emerge. A web application interface accompanies the system, providing an intuitive means of interacting with the service. The implementation confirms the feasibility of employing agent-based LLM architectures in the automotive domain, delivering accurate responses to user queries. The system establishes a foundation for future enhancements and demonstrates the potential of LLM-driven multi-agent systems across diverse application domains.