Responsible Tutor and Live Chatbots serving education at EPFL

At Applied Machine Learning Days 2026, we led a workshop presenting the progress of our project to integrate Tutor Bots into EPFL course forums and course-specific Live Chatbots.

Our presentation was based on the Digital Ethics Canvas, a tool developed jointly by EPFL and the University of Neuchâtel. This framework allows socio-technical choices to be analyzed and structured in light of explicit ethical and pedagogical principles. Workshop participants were thus able to explore the concrete decisions made during development and the associated trade-offs.

Choices guided by key values

Our approach is based on core values of responsibility:

• Data protection and confidentiality: Data remains within EPFL thanks to the internal infrastructure provided by RCP. We use open-weight language models that operate locally.
• Fair access: All students in a course have the same opportunities to interact with the bots.
• Pedagogical alignment: Depending on the preferences of the teachers, the bots can provide clues rather than direct answers, in order to promote autonomy and active learning.
• Accuracy and human supervision: We have developed a RAG (Retrieval-Augmented Generation) pipeline enriched with filters and automated intermediate evaluations. Teachers can also choose whether responses posted on forums appear immediately or only after validation. In all cases, the teaching team remains solely responsible for ensuring the quality of responses.
• Digital sobriety: We favor lighter models that require less memory and energy resources than large commercial solutions.

Phase 2 of the pilot

The spring semester marks phase 2 of our pilot study. Among the improvements developed is the integration of content posted on Moodle into our RAG pipeline, which enhances the usability of our system for teachers.

We currently support more than 20 courses (including MOOCs), with the aim of continuing to grow over the coming semesters. Our goal remains unchanged: to improve the system iteratively, in line with our core values and in the service of responsible use of AI in education.