Mid Sweden University

The Draft-Based Iterative Feedback Accelerator (DIFA) is an AI-enhanced extension of the FAMS (Shahzad et al., 2023), which is designed to address challenges in providing scalable, high-quality formative feedback in higher education. Formative feedback is crucial for metacognitive development and self-regulated learning, yet its effective implementation is often limited in large cohorts, particularly for neurodiverse learners who need structured and transparent feedback. Moreover, the rise in students' reliance on generative AI raises concerns about deep learning, highlighting the need for assessment designs that promote meaningful cognitive engagement. DIFA transforms AI from a passive tool into an active pedagogical assistant within the feedback loop. It allows students to submit ongoing, incomplete work for evaluation, shifting focus from outcome-oriented grading to a process centered on reflection and continuous improvement. DIFA features reusable feedback fragments and a repository of common errors.

DIFA offers a modern approach to feedback by processing student drafts to generate context-specific responses that instructors can refine. This makes iterative feedback cycles more manageable within existing workloads. It enhances the Feedback and Assessment Management System (FAMS) by using natural language processing (NLP) to automate the categorization of feedback and create descriptive titles, improving organization and retrieval. When predefined feedback is unavailable, a lightweight language model trained on historical data generates relevant and clear responses, transforming the system into an adaptive feedback generator.

The architecture employs a teacher-in-the-loop approach where AI-generated feedback serves as a draft for instructors to review and refine. This ensures that AI enhances, rather than replaces, educators' professional judgment. By integrating AI into organizing and refining feedback, the DIFA system creates a continuous, data-driven feedback cycle that lightens instructors' workloads while delivering timely, personalized responses. Moreover, incorporating AI within a structured assessment framework helps prevent students' misuse of generative AI, positioning it as a tool for genuine learning.

Preliminary results show that this AI-enhanced architecture improves feedback delivery while maintaining the necessary depth and personalization for effective formative assessment. The DIFA system allows for iterative revisions rather than just final outputs, illustrating how explainable AI can enhance educational expertise. Future developments will include a student-facing interface that focuses on adaptive explanations and interactive feedback for neurodiverse learners

Virtual learning environments offer new possibilities for technology enhanced teaching and learning, but providing rapid, individualised feedback for complex assignments in large student groups remains challenging. This paper presents a Formative Assessment Management System (FAMS), a computer-based tool for teachers to generate written feedback at scale with minimal overhead. FAMS leverages archived feedback fragments and thematic identifiers to create pertinent feedback while consistently maintaining quality and fairness. The system has been implemented in programming courses and yielded promising results, including reduced feedback delivery time and maintained feedback quality. Future research will evaluate FAMS from student and teacher perspectives, conforming to educational action research, continuous quality improvements, and investigating correlations between aspect-based assessment and learning outcomes.