Iris — TUM's Hint-Giving AI Virtual Tutor Inside the Open-Source Artemis Platform for Programming Courses

Iris was developed by Patrick Bassner, Eduard Frankford and Stephan Krusche at the Chair of Software Engineering, Technical University of Munich, and integrated into the open-source Artemis interactive learning platform in October 2023. The ITiCSE 2024 paper describes an implementation based on GPT-3.5-Turbo with chain-of-thought prompting, tutor role descriptions and few-shot examples. It reads the problem statement, the student's code and automated test feedback, and deliberately avoids giving complete solutions, offering subtle hints or counter-questions instead. It answers only questions relevant to the course.
The evaluation surveyed over 200 students across three introductory computer-science lectures. Students said Iris improved their understanding of programming concepts, gave personalised support that complemented human tutors, and offered a judgement-free space to ask questions; they reported feeling confident about exams without it. Students also valued that their chats remain private.
The evidence is descriptive: it is based on student perceptions rather than a controlled comparison of learning outcomes, and the study is from one university. Its value lies in a transparent design that puts pedagogy before answer-giving, which contrasts with the engagement losses seen in unguided tutor deployments elsewhere. The developers planned further optimisation and extension to other subjects.

Read the full analysis: https://www.sciencedaily.com/releases/2024/07/240717120057.htm