ChAx: a RAG-based chatbot for CAx education
Sarah Steininger, Saltuk Kezer, Jona Rief, Emily Spicker, Sebastian Preis, Johannes Fottner · Proceedings of the Design Society · 2025
ABSTRACT: ChAx is a chatbot designed to support in technical drawing lectures by leveraging Retrieval-Augmented Generation. Addressing challenges such as the complexity of rules and dependencies in technical drawing, the system accesses the specific lecture materials to provide students with accurate and context-aware answers. The architecture combines modular components, including a RAG pipeline and a frontend with an interactive PDF viewer, ensuring transparency and user-friendliness. Optimization strategies like semantic chunking, fine-tuning, and cost-effective configurations enable efficient performance within constrained server environments. Evaluation metrics, including factual correctness and answer relevancy, were evaluated by using the LLM-as-a-judge method. The results underline ChAx’s potential to enhance educational outcomes by enabling students utilize materials more effectively.