GAIMHE

Generative AI Models for Hybrid Educational technologies

Overview

GAIMHE is a strategic project funded by Bpifrance to build the future of how AI serves education. This collaboration is coordinated by EvidenceB and brings together the Flowers AI & CogSci Lab at Inria, Café pédagogique, and Association Class’Code to create the next generation of intelligent learning systems.

💡 THE CHALLENGE

Today’s educational AI faces a critical trade-off:

1️⃣ Traditional Intelligent Tutoring Systems (ITS) like AdaptivMaths or MIA Seconde deliver pedagogically grounded, personalized curricula using efficient, cognitive science-inspired AI algorithms (like the ZPDES AI algorithm developed in the Flowers AI & CogSci Lab). But they require extensive manual content creation at the start, e.g. to populate the graph of exercices.

2️⃣ Generative AI offers flexibility and natural interaction, but systems including the learning modes of the main genAI softwares lack pedagogical rigor (they can even encode harmful pedagogical biases), can’t personalize year-long curricula, and consume massive energy.

🎯 OUR APPROACH

GAIMHE bridges this gap by hybridizing both paradigms:

1️⃣ GenAI-powered content pre-generation to automatically create pedagogically compliant and diverse exercises, rapidly populating ITS graphs 2️⃣ Targeted generative guidance that applies proven pedagogical principles only when needed, using optimally-sized models 3️⃣ Next-generation personalization through compact student models that predict and adapt to learning trajectories across vast exercise spaces (building on our MAGELLAN research)

🌍 REAL-WORLD IMPACT AT SCALE

This isn’t only lab research: we aim for actual deployed innovation. EvidenceB’s platforms already serve tens of thousands of classrooms across primary, middle, and high schools in multiple disciplines (AdaptivMaths, MIA Seconde, etc). GAIMHE will leverage this massive real-world infrastructure, learning from real educational data and evaluating outcomes where it matters: actual classrooms.

♻️ DIGITAL COMMONS

To accelerate France’s EDTech France ecosystem, we will release major project components as open-source and digital commons: annotated datasets, learning traces, and software (both genAI and learning analytics over datasets like AdaptivMaths). This commitment is reinforced through our partnership with Région Ile de France, leveraging their digital commons textbooks.

Partners

EvidenceB • Inria • ClassCode / OpenLLM-France • Région Île-de-France • Le café pédagogique • Scaleway