Pierre Colombo, PhD

Research scientist at Google DeepMind. I work on Gemini for the law.

Before that I trained multimodal models at Cohere, and co-founded Equall, where we built LLMs and agents for M&A due diligence and put them in the hands of large law firms. Before that I was an Associate Professor (maître de conférences) at CentraleSupélec, where I started the NLP group.

What I built

• LawSaulLM. The first LLMs built for law.
• RetrievalColPali, ModernVBERT, BidirLM. Multimodal retrieval: text, images and documents.
• LLMsEuroLLM, EuroBERT, CroissantLLM, TowerLLM. Europe’s open multilingual models.
• EvaluationInfoLM, xCOMET. Metrics for generated text and translation. InfoLM got an AAAI 2022 paper award.

The rest is on Scholar.

PhD students I advised

As an Associate Professor I advised six PhDs: Manuel Faysse (Google DeepMind), Maxime Darrin (Mistral), Nicolas Boizard (Cohere), Duarte Alves (Reflection), Hippolyte Gisserot-Boukhlef (QRT) and Nuno Guerreiro (Sword).