About AI-D-Mentor
AI Diagnostic Mentor: a tutor that finds out what a student actually understands about a course, and tells both the student and the teacher.
A general chatbot will answer anything, fluently, and leaves no comparable record of what a student knows. AI-D-Mentor works the other way round. It starts from questions the teacher wrote, stays inside the course, judges each answer against the course's own lecture material, and keeps asking within a topic until it has enough evidence. Every session is stored so the teaching team can review it.
A session, step by step
- The student signs in with a four-character code from their teacher. The code is all the system stores; which student holds which code stays with the teacher.
- The mentor asks a question the teacher wrote — open, multiple choice or true/false. A student can attach a photo of a hand-drawn diagram or derivation to an answer.
- The answer is graded against the lecture slides and transcripts: which of the required points it covers, not how long or polished it is. "I'm not sure" is an accepted answer.
- Follow-up questions probe whatever the answer left open, within the same topic.
- When a topic is done, the student sees the correct answer as a short checklist, with references to the slides it comes from.
A session has a time limit the teacher sets, and the teacher can extend it or close it for the whole class while it is running. A teacher can also offer a practice mode: no clock, the answer shown after every question, and no grade.
What teachers get
- Course set-up from the material. The mentor reads the lecture decks and proposes the course description, the learning objectives and draft questions with model answers. The teacher edits and approves; nothing reaches students unapproved.
- Lecture recordings as material. Recordings can be turned into checked transcripts, so what was said in a lecture counts, not only what was on the slides.
- Review. Every session, answer by answer, with the grade, the reasoning and the material it was graded against. Teaching assistants can review too, and their corrections are used to calibrate the grading.
- Results per topic for each student and for the class, showing where it is strong and where it is not — useful for the next lecture as much as for assessment.
Data and copyright
- Students are anonymous to the system: a code, no name, no email.
- Everything runs on this Chalmers virtual machine. The grading model is reached through the Chalmers AI gateway.
- Slides the teacher marks as copyright-restricted stay on the server and are not sent to the model.
Status
AI-D-Mentor was first used with students at the IDEA League summer school Analysis and Modelling of Road-User Behaviour at Chalmers in August 2026, and has been developed further from what that cohort showed. It is a research and teaching tool under active development, not a Chalmers-wide service.
Try it yourself: the demo is a made-up course you can take as a student, or open as its teacher or teaching assistant — no account needed. To use it on your own course, write to jonas.bargman@chalmers.se.