Language learning theories have long shaped the design of interactive CALL systems. For example, many of the earliest platforms provided both multiple choice quizzes around comprehensible input and gap fill tasks promoting output. More recently, the emergence of generative AI has led to debates about whether a CALL system could be a human-like learning partner.

This poster demonstrates how fine-tuning through ChatGPT's API can be carried out to create an AI-based interactive CALL system for real-time feedback on learners’ academic writing. Fine-tuning is carried out based on teacher’s insights into students' academic writing capabilities and writing needs, here demonstrated with ICNALE corpus data. The AI is integrated into a CALL system designed to facilitate in-class discussions on academic writing.

Fine-tuning allows the system to reflect the teacher’s understanding of their students' needs, improving its responsiveness and feedback quality. The goal of this work is to demonstrate how current AI systems can be rewired using teachers' insights to create an environment where AI not only provides feedback but also supports classroom-based discussions on academic writing. This is a useful demonstration towards exploring more specialized, responsive systems.