Engineer H/F: Post-Training Generative Models for Education at Scale

Descrizione

Testo originale dell'annuncio (in inglese)

The Flowers AI & CogSci Lab at Inria, in partnership with EvidenceB, Café pédagogique, and ClassCode, is launching GAIMHE (Generative AI for Hybrid Mathematics Education), a large-scale research and innovation project funded by Bpifrance. This initiative addresses a critical challenge in educational technology: developing AI systems that combine the pedagogical rigor and personalization capabilities of Intelligent Tutoring Systems (ITS) with the flexibility and generative power of modern large language models. Current ITS platforms, such as EvidenceB's AdaptivMaths, leverage cognitive science principles and structured pedagogical graphs to deliver personalized learning pathways to students. These systems have demonstrated effectiveness across tens of thousands of classrooms in France (primary, middle, and high schools, across multiple disciplines including AdaptivMaths and MIA Seconde). However, their development requires substantial manual content creation. Conversely, generative AI offers unprecedented flexibility but often lacks pedagogical grounding, cannot sustain long-term curriculum personalization, and raises concerns about energy efficiency and pedagogical biases. GAIMHE will develop hybrid architectures that harness generative AI for automated content generation while maintaining pedagogical constraints, deploy targeted generative guidance aligned with established learning theories, and create compact student models for next-generation personalization algorithms. The project will leverage EvidenceB's extensive deployment infrastructure to work with authentic large-scale educational data and validate innovations in real classroom settings. In alignment with open science principles and through partnership with Région Île-de-France, major project outputs (datasets, models, software) will be released as digital commons under open-source licenses. Design, implement, and evaluate generative AI systems for automated creation of pedagogically compliant educational exercises and content Develop and optimize agentic architectures integrating large language models with structured ITS frameworks, ensuring pedagogical alignment and computational efficiency Implement and fine-tune small-scale generative models for student learning trajectory prediction and personalized curriculum adaptation Deploy LLM-as-judge frameworks and reinforcement learning approaches to evaluate and improve pedagogical quality of AI-generated content Conduct large-scale experiments analyzing learning traces and student interactions with hybrid AI systems in authentic classroom environments Collaborate with pedagogical experts, cognitive scientists, and industrial partners to translate educational requirements into technical specifications Contribute to open-source software development and documentation for digital commons dissemination Participate in scientific valorization through publications, presentations, and technical reports Full description: https://www.pyoudeyer.com/researchEngineerGAIMHE26.pdf