TOYOTA R&D ASSISTANT: A LLMs based Multi-Agent System for Conversational Battery Data Analysis

The rapid increase in data generation across sectors like automotive manufacturing—driven by the rise of electric vehicles and resulting in unprecedented volumes of data, particularly in areas such as battery testing—has underscored an urgent need for efficient and user-friendly tools for complex data analysis. Traditional methods, including SQL, Python-based workflows or tools like Microsoft Excel, require advanced technical expertise, limiting their adoption among non-specialists. To address this challenge, this thesis proposes a novel multi-agent virtual assistant powered by Large Language Models (LLMs), designed to democratize data analysis through conversational interaction. The developed system supports natural language interaction to enable data extraction, processing, and the construction of custom visualizations, thereby reducing the need for specialized coding skills. Its modular architecture allows seamless adaptation to diverse data types and domains, with a primary focus on battery performance analytics. By bridging the gap between technical and non-technical users, the solution enhances productivity, improves analytical accuracy, and offers a scalable framework for cross-sector data-driven decision-making. Experimental results demonstrate its effectiveness in reducing analysis time while maintaining robustness across heterogeneous requests, positioning it as a versatile tool for industries facing similar data scalability challenges.