Vincent Michel

Technology leader & builder | AI, Data & Software | From exploration to strategic scale

I like building things before they become obvious — exploring technical opportunities hands-on, proving what works, then building the teams and systems that turn them into strategic capabilities.

I currently serve as General Manager of Recommendations within Rakuten’s AI Data Division, leading international teams building and operating large-scale recommendation systems.

At Rakuten, I helped turn recommendation from a small team running relatively simple systems into a multi-team engineering capability, while scaling served traffic by more than an order of magnitude. I built some of the early systems myself, then progressively moved from engineering and ML into architecture, product, team building, investment decisions and strategy — helping establish recommendation as a visible and trusted capability within the company.

My approach to technology is deliberately end-to-end. I like understanding a problem deeply enough to move across research, algorithms, software, infrastructure, operations and product, identify where the real leverage lies, and bring together the right expertise to take it further. As projects mature, my role naturally shifts from building the first version to creating the environment, organization and executive support needed to scale them.

Earlier in my career, I trained as an engineer at ESPCI Paris, one of France’s highly selective grandes écoles, and worked in machine-learning research at INRIA, Université Paris-Sud and NeuroSpin. In 2010, I was part of the INRIA team that took leadership of scikit-learn and delivered its first public release [1]. It has since become one of the reference open-source libraries in machine learning.

Beyond scikit-learn, my research focused on machine learning for high-dimensional neuroimaging, with peer-reviewed work including Total Variation regularization for predictive brain imaging in IEEE Transactions on Medical Imaging [2] and supervised clustering in Pattern Recognition [3].

I later worked at Logilab on software and data systems, including data.bnf.fr, the French National Library’s linked open-data platform, recognized with the Stanford Prize for Innovation in Research Libraries [4].

I have also been involved in building the recommender-systems community in France, notably as a co-organizer of RecSysFR from 2016 to 2018. The Paris meetup had grown to more than 700 members during that period and has since reached around 950 members [5].

Today, my interests extend well beyond AI as a field in itself. I am particularly drawn to situations where software, data and AI can create or transform a real product or industrial capability — especially when the opportunity is still ambiguous, the path is not obvious, and there is something meaningful to explore, build and scale.

My work increasingly sits at the level of technology direction, organizational design and strategic investment: deciding where to place bets, building the teams and capabilities to pursue them, and creating the executive support needed to turn promising ideas into lasting outcomes — while staying close enough to engineering and product to test ideas, challenge technical choices and understand what is really happening.

References

[1] scikit-learn — Official project history
[2] Michel et al. — Total Variation Regularization for fMRI-Based Prediction of Behavior, IEEE Transactions on Medical Imaging, 2011
[3] Michel et al. — A supervised clustering approach for fMRI-based inference of brain states, Pattern Recognition, 2012
[4] Bibliothèque nationale de France — data.bnf.fr / Stanford Prize for Innovation in Research Libraries
[5] RecSysFR — Paris Recommender Systems Meetup