AI Tutors in Ed Discussion Forums: Supporting Teaching and Learning

The CEDE team has recently launched a pilot introducing AI tutors to support teaching teams in answering students’ questions within Ed discussion forums. These tutors are powered by locally hosted open-weight LLMs, ensuring that all data remains securely within EPFL’s infrastructure.

We have designed the system with two dimensions of customization:

• Interaction mode
  • Adventurous: the bot provides a direct response to students, which will be reviewed a posteriori by the pedagogical team.
  • Safe: the bot drafts a response that is first reviewed by the pedagogical team before being released to students.
• Tutoring modality
  • Direct: the bot provides a more direct and complete response.
  • Hint-based: the bot provides guidance and hints.

Each professor can select the combination that best aligns with their pedagogical approach.

So far, eight AI tutors have been deployed across courses at EPFL:

• MICRO-331 Microfabrication technologies (Jürgen Brugger)
• PHYS-101(l) General Physics Mechanics (Cécile Hébert)
• MATH-101(e) Analyse I (Anna Lachowska)
• MATH-310 Algebra (Anna Lachowska)
• MATH-111(c) Linear Algebra (Orane Pouchon)
• CH-314 Structural Analysis (Pascal Miéville)
• CS-119(d) Information, Calcul, Communication (Jean-Cédric Chappellier)
• PHYS-201(a) General Physics: Electromagnetism (Joaquim Loizu)

We thank the volunteers for their trust and openness to innovation. An additional six tutors are currently being prepared.

This initiative is part of our broader effort to explore retrieval-augmented generation (RAG) technologies for higher education. By combining local large language models with course-specific materials, we aim to reduce the risk of hallucinations and to provide students with consistent, context-aware responses.

The deployment of these AI tutors is a collaborative effort of the CEDE team: Patrick Jermann, Javier Lloret Pardo, Florian Dufour, Annechien Helsdingen, Ramtin Yazdanian, and Aitor Pérez with the help and support from Farah Elsousy, Koami Gafan and Nicolas Teissier.