FEEDBOT: Formative Design Feedback on Programming Assignments
Elaine Zhu, Smaran Teja, Chris Coombes, DANIEL C. PATTERSON · 2025
This paper describes FEEDBOT, an open-source formative assessment tool leveraging large language models (LLMs) to provide structured, high-level feedback on design-oriented programming assignments. Designed to address the limitations of traditional autograding and overcome scaling challenges of formative assessment, FEEDBOT uses an existing pedagogical framework to provide targeted but limited actionable feedback. Early results demonstrate measurable improvements in student performance in a large introductory computer science course, while avoiding the pitfall of providing too much assistance that more unstructured tools commonly encounter. Although this experience report focuses on a particular implementation, we believe that FEEDBOT (and its general approach) is adaptable to many other contexts.