OCTAVIA : Òc to intelligence, No to dependence

ÒCTAVIA

Borde Mathieu, Da Silva Olivier, Goujon Maël, Lizy-Pousson Youri, André Aoun(mentor) from University of Toulouse

Training / education / pedagogy

Students currently use generic generative AIs passively, treating them as mere "answer machines" which bypasses the learning process. They also face heavy cognitive and organizational overload. To solve this, we are working on a sovereign, multi-agent AI ecosystem integrated into the university infrastructure. Our use case involves three AI agents collaborating exclusively for the student: a Planner that manages complex schedules, a Content agent that transforms official course PDFs into interactive tools (quizzes, mind maps), and " ÒCTAVIA," a tutoring agent. Using a strict RAG system on verified course data, ÒCTAVIA never gives direct answers, but uses guided questions to force students' critical thinking.

Current AIs create a severe cognitive tension: by providing immediate solutions, they foster intellectual passivity rather than deep learning. This project matters because it transforms AI into a true pedagogical partner by targeting major student pain points. First, it alleviates the mental load of organization through the Planner. Second, it counters the passive use of AI by forcing reasoning through the ÒCTAVIA agent. Finally, generating personalized study sheets and quizzes caters to diverse learning styles, promoting educational equity. Furthermore, generic LLMs raise critical data privacy issues. Our sovereign approach ensures every student accesses a secure tool based exclusively on their professors' validated materials.

Our primary motivation is to genuinely change students' daily lives by providing them with a tool that brings real value and relieves their academic burden. This is encapsulated in our slogan: "Oc to intelligence, no to dependence." We want to shift AI from a tool that bypasses education to one that actively empowers the student. As a team, our learning objectives focus on upskilling in agentic AI and multi-agent orchestration. We aim to learn how to design complex, interacting AI systems and how to integrate a trusted, sovereign European LLM to guarantee data privacy, prioritizing pedagogical impact over simple task execution.

Here is the link to access our initial contribution in PDF format: ⠀ https://drive.google.com/file/d/1ZaiEP78f88EJIZorYi4B3hauLYzVkoBm/view?usp=sharing