Non-experts often struggle to interpret and apply unstructured feedback, whether it is provided by human experts or Large Language Models (LLMs). They also find it challenging to map feedback to high-level principles essential for effective learning.

To address this challenge, we present Feedstack, an interactive system designed to bridge this gap by creating bidirectional mappings between direct feedback and abstract visual design principles. Feedstack's design is informed by two formative studies (N = 17), including semi-structured interviews and observations that explored how students request and receive feedback using large language models (LLMs). These two formative studies directly informed our system's design. We evaluated the Feedstack prototype through a user study (N = 8), where participants engaged with the system and reflected on its ability to structure feedback. Post-study surveys and interviews offered additional insights into how structured feedback, informed by high-level design principles, affected their experiences and feedback exchange. Our contributions include a conceptual framework and novel affordances for feedback structuring, aimed at helping non-experts better understand and apply high-level principles consistently across lengthy unstructured feedback conversations. This research contributes to advancing interactive feedback systems, particularly in educational contexts where real-time design feedback is critical for learning and skill development.