Research & Papers

TeachingCoach: A Fine-Tuned Scaffolding Chatbot for Instructional Guidance to Instructors

Researchers' specialized model provides clearer, more reflective teaching advice than general-purpose AI.

Deep Dive

A research team led by Isabel Molnar has developed TeachingCoach, a specialized AI chatbot designed to provide real-time, pedagogically grounded guidance to higher education instructors. The system addresses a critical gap in scalable instructional support, moving beyond generic chatbot advice or limited human consultations. TeachingCoach employs a novel data-centric pipeline that extracts pedagogical rules from educational resources and uses synthetic dialogue generation to fine-tune a language model specifically for instructional coaching.

The model guides instructors through a structured process of problem identification, diagnosis, and strategy development. In expert evaluations, TeachingCoach produced clearer, more reflective, and more responsive guidance compared to a GPT-4o mini baseline. A user study with actual higher education instructors revealed important trade-offs between conversational depth and interaction efficiency, highlighting both the system's strengths and areas for improvement in practical deployment scenarios.

This research demonstrates that pedagogically grounded, synthetic data-driven chatbots can significantly improve instructional support systems. The approach offers a scalable design framework for future educational AI tools, showing how domain-specific fine-tuning can create more effective solutions than general-purpose language models. The work represents an important step toward AI systems that understand and support professional development in specialized fields like education.

Key Points
  • Specialized fine-tuning outperforms GPT-4o mini in expert evaluations for clarity and responsiveness
  • Uses synthetic dialogue generation from extracted pedagogical rules to create domain-specific training data
  • User study reveals trade-offs between conversational depth and interaction efficiency in real-world use

Why It Matters

Shows how domain-specific AI fine-tuning creates more effective professional tools than general-purpose models, with implications across specialized fields.