Stage 2027 AI/ML Engineer 100% R&D - Tabular foundation models
Overview
Ekimetrics is offering a six-month research and development internship titled Stage 2027 AI/ML Engineer 100% R&D - Tabular foundation models, located in Paris, France. The intern will work within the Innovation Department alongside senior researchers and data scientists on an emerging research topic at the intersection of tabular modeling, latent reasoning, and interpretability. The position focuses on designing and evaluating latent reasoning mechanisms for Tabular Foundation Models and applying interpretability methods to analyze their computations.
Who can apply
Candidates must currently be in the final year of a Master's or Engineering degree with strong foundations in machine learning, statistics, and programming. Applicants should possess a good grasp of model validation, strong analytical skills, and the ability to work autonomously on an open-ended research problem. While not mandatory, knowledge of transformers and language models, familiarity with latent reasoning or recurrent-depth models, experience in interpretability, knowledge of TabPFN, TabICL, or TabDPT, solid software engineering practices, and fluency in English are considered advantages.
What you get
The internship provides hands-on research and engineering experience, collaborative learning with a team of data scientists and PhD experts, and skill development in AI system evaluation. Participants have opportunities for career growth including potential transition into a full-time Data Scientist role in AI at Ekimetrics or into a PhD program upon successful completion. Additional benefits include the possibility to co-author and publish scientific papers in reputable AI and data science conferences and journals, networking with industrial and academic partners, and experience in ethical and sustainable AI development.
Requirements
The role requires strong Python and PyTorch skills, alongside a solid command of the standard data science toolkit including pandas, polars, scikit-learn, and gradient-boosting libraries. The intern will be responsible for reviewing the literature on latent reasoning and tabular foundation models, identifying and adapting latent reasoning mechanisms to TFM architectures, and implementing them on open-source TFMs. Responsibilities also include measuring effects on accuracy, calibration, and robustness against baselines, applying interpretability methods such as representation analysis, probing, and activation patching, building an evaluation protocol on diverse tabular datasets, documenting results, and contributing to scientific publications.
Key dates
The duration of the internship is six months.