Selected resources

Slides, posters, and materials from my research — free to read and share.

PhD defense

Guiding the minds of tomorrow: conversational agents to train curiosity and metacognition in young learners

Defense slides · Université de Bordeaux / Inria · defended September 3, 2024

My PhD asked whether educational technologies can train — rather than replace — children's curiosity. I designed conversational agents that guide 9–11-year-olds through the practice of curious question-asking and the metacognitive skills behind it, and evaluated them in a series of classroom studies in French primary schools. The results show that practicing these skills with well-designed agents supports children's divergent question-asking and their autonomous, curiosity-driven learning — and the thesis proposes design principles for building such technologies.

Watch the defense

Plays on YouTube (privacy-enhanced mode) · watch on YouTube ↗

Selected conference posters

From prompting to epistemic proactivity: temporal trajectories of student–AI interaction in mathematics learning

Poster · AIED 2026, Seoul · with Peter Kaiser & Kou Murayama

Can students stay epistemically proactive when an AI tutor will answer anything? In this study, 112 German secondary-school students (ages 14–15) practiced mathematical modelling with an LLM tutor while every conversational turn was coded — with LLM-assisted, human-validated coding — for self-regulation, help-seeking, and modelling activity. Students typically began with high-quality help-seeking but drifted toward executive answer-seeking over the session, and these temporal shifts predicted their independent post-test performance. The takeaway: how student–AI interactions evolve over time, not just their average quality, is what matters for learning.

The illusion of understanding: how middle-school students fail to regulate inquiry with ChatGPT during a science task

Poster · AIED 2026 · with Kou Murayama, Celeste Kidd, Hélène Sauzéon & Pierre-Yves Oudeyer

Effortful inquiry — formulating precise questions, evaluating the answers, following up — is central to science learning, but generative AI makes it easy to skip. We had 63 middle-school students (ages 14–15) work through science tasks with ChatGPT and measured their sensitivity to prompt quality, their evaluation of AI answers, and their follow-up questioning. On average, students barely distinguished good from poor prompts or answers, and rarely asked clarifying questions after low-quality responses. The results point to AI literacy and metacognitive training as prerequisites for productive AI-supported inquiry.

Educational videos

Curiosity educational videos

Video series · KidsReflect project

A set of short educational videos designed to introduce children to curiosity: what it is, why it matters for learning, and how to practice curious question-asking in the classroom.