FeedFix: Generating Abstract and Personalized Feedback for Introductory Programming Assignments Through Automated Program Repair

October 27, 2025

The increasing number of students that enroll in higher education programming courses, along with the significant increase in the availability of Massive Online Open Courses, have made it nearly impossible for personalized feedback to be given to students who are now starting to learn programming. This reduction in the amount of personalized feedback leads to frustration and a slower rate of learning from the students, whose work and progression suffers from the lack of guidance. In this Thesis, we present FeedFix, a tool designed to tackle this problem, by taking advantage of Fault Localization to pinpoint errors in the program and leveraging the recent improvements in the realm of Large Language Models, when it comes to Automated Program Repair, to create personalized solutions based on the students’ own buggy submissions. FeedFix then takes these solutions and its corresponding buggy student submission to generate a detailed Sketch of the solution, revealing the structure of the program, while abstracting away any specific portions of code of the student submission that were incorrect or missing. FeedFix also allows the faculty members to alter the level of feedback given, giving them the freedom to provide more or less detail in the sketch, as they see fit. With this in mind and considering that FeedFix’s best configurations have over a 50% repair rate, FeedFix shows tremendous promise when it comes to tackling this feeedback problem.

Authors in the community:

• 

  T

  Tiago Miguel Rodrigues da Costa e Silva Vaz

  ist196913

Supervisors of this institution:

• 

  V

  Vasco Miguel Gomes Nunes Manquinho

  ist14221

• 

  P

  Pedro Orvalho

  ist181151

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

eng - English

Instituto Superior Técnico