This thesis looks at how large language models (LLMs) can be used to improve customer service and help smart cities reach their sustainability goals, with a focus on e-mobility infrastructure.

This study uses a design science approach to create and test a dynamic frequently asked questions (FAQ) system powered by LLMs to offer real-time, context-aware customer support and get around the problems with static information systems. The development process involved fine-tuning the LLM (Llama) with public datasets, proprietary knowledge from ewl (Energie Wasser Luzern), and structured customer service inquiries. A hybrid rule-based system was implemented to combine Llama with Voiceflow, enhancing the system's ability to manage both general and domain-specific queries. Evaluation included testing four models - Llama, Voiceflow, a hybrid rule-based model, and CoPilot Pro - against criteria such as accuracy, relevance, and clarity. Expert feedback highlighted the strengths of LLM-based systems in improving user satisfaction and accessibility while revealing challenges such as resource constraints and reliance on high-quality data.

This research shows how LLMs can tackle sustainability challenges in AI and smart cities by delivering customized information. It also lays the groundwork for future applications, including scalability to other urban domains and ethical considerations like data privacy.