Entwicklung eines KI-Chatbots als Lernpartner für die Vorlesung Kfz-Haftpflichtversicherung auf Basis von Retrieval-Augmented Generation (RAG)

A retrieval‑augmented generation (RAG)‑based chatbot was created to serve as a learning companion for a university course on motor‑vehicle liability insurance, delivering answers that are directly grounded in the lecture materials and citing the exact source pages. Three interaction modes (chat, quiz, and sparring) were implemented, employing a multi‑query extension, hybrid semantic‑lexical search, rank fusion and multilingual reranking, with the optimal configuration identified through evaluation on a 40‑question test set using RAGAS metrics. The system’s strengths, limitations (e.g., response time and reliance on a commercial embedding service) and possible extensions such as fully local deployment and empirical studies of learning outcomes were discussed.