Smarter Coaching Chatbots: Leveraging Hybrid Architectures for Deeper Context Awareness

This paper examines the integration of Large Language Models into digital coaching for higher education students, using the StudiCoachBot as a case study. A user study involving 83 participants was conducted to assess the influence of LLM-supported chatbot coaching on working alliance, social presence and user acceptance. The practical insights gained informed the design of a hybrid chatbot architecture blending rule-based dialogue guidance with a meta-assistant that can select context-appropriate, LLM-powered modules for specific coaching situations. This hybrid model aims to ensure transparency and structure while providing personalized, responsive support. Our findings reveal opportunities and tensions associated with the use of generative AI for student coaching, suggesting future directions for the responsible and meaningful provision of digital support.