Collaborations

Current collaborations

• LightOn — work on understanding extremely large language models, specialized multimodal model building and evaluation. (BPI Scribe + CIFRE Thesis, Maxence Lasbordes)
• Meta - Work on Agentic LLMs, reasoning, planning. (CIFRE Thesis, Romain Froger)
• Thales - Work on LLM Safety, focus on intentional biases transmission. (CIFRE Thesis, Lisa Bouger)
• Dr. Tatjana Scheffler with Melce Husunbeyi (PhD student). Work on multimodal fact checking, inclusion of knowlege graph into discriminative LLMs.
• Pr. Michelangelo Ceci with Francesco Benedetti (PhD student). Cross-domain online radicalisation detection, syntethic network generation, socio-demographic biases analysis.

Past collaborations

• AXA Group / AXA Research — bertology and interpretability for black-box NLP models.
• ReciTAL.ai — synthetic data augmentation for multilingual question answering.
• QuantCube / Q3 Technology — multilingual NLP, fact-checking, and information quality evaluation.
• Fortia Financial Services, now Outscale — preprocessing pipelines and evaluation datasets for semantic analysis.
• Davi — French NLP parsing pipelines and unsupervised clustering for domain adaptation.