Research Scientist - Machine Learning - Foundation Models H/F

collectivite

Information importante

Type de contrat: CDI

Salaire : 70-100K fixe

Localisation : Paris, France

Date de démarrage :

4 à 8 semaines

Mode de travail : Hybride

Publié le : 2 juillet 2026

Le besoin

Company

Our client is an ambitious AI deeptech company developing a new generation of foundation models for structured and tabular data.

The company is building proprietary AI models designed to represent complex structured datasets, improve transfer learning capabilities, and turn data into actionable insights for strategic industrial applications.

The team is still small, highly scientific, and backed by strong ambitions. This is an opportunity to join a company working on a technically rare and intellectually demanding field, at the intersection of machine learning research, data representation, and large-scale model training.

Role Overview

As an AI Research Scientist, you will contribute directly to the design, training, evaluation, and improvement of in-house AI models for structured data representation.

This role is primarily research-oriented. The company is looking for someone with strong depth in machine learning, deep learning, Transformers, and model training from scratch.

You will work closely with the research and engineering teams to improve the performance, scalability, robustness, and adaptability of the company's AI models. You may also collaborate with industrial and academic partners to translate research progress into practical, high-impact applications.

Key Responsibilities

Model Research & Architecture

Contribute to the design and development of new machine learning architectures for structured and tabular data representation.

Work on embedding models, representation learning, transfer learning, and foundation model approaches applied to structured datasets.

Explore new learning paradigms and architectural choices to improve model performance, scalability, and generalization.

Training & Pretraining

Design and run training experiments for deep learning models, including models trained from scratch.

Contribute to pretraining strategies, training data optimization, and large-scale experimentation.

Work on improving training pipelines, sample selection, dataset construction, and evaluation protocols to strengthen the representational capabilities of the models.

Evaluation & Analysis

Continuously evaluate model performance using adapted metrics aligned with research objectives and real-world use cases.

Run ad-hoc analyses to better understand model behavior, learning mechanisms, strengths, and limitations.

Contribute to the design of robust evaluation frameworks for downstream applications.

Research & Scientific Contribution

Stay up to date with the latest advances in machine learning, deep learning, Transformers, representation learning, and foundation models.

Translate relevant research ideas into concrete experiments and model improvements.

Communicate research concepts clearly to both technical and scientific audiences.

Collaboration

Work closely with ML researchers, ML engineers, data scientists, and external partners.

Collaborate with the engineering team to ensure research outputs can be scaled, reproduced, and integrated into production-oriented workflows when relevant.

Must-Have Requirements

PhD in Computer Science, Machine Learning, Deep Learning, Artificial Intelligence, or a closely related field.

Strong expertise in Machine Learning, Deep Learning, and Transformers.

Real experience designing model architectures and training models from scratch.

Experience with pretraining is highly preferred.

Strong understanding of modern ML research practices, including experimentation, reproducibility, scalability, and evaluation.

Publication record in top-tier ML conferences or journals, directly related to the targeted research topics.

Hands-on experience with deep learning frameworks such as PyTorch, JAX, or similar.

Experience running large-scale ML experiments in cloud environments or private clusters.

Strong analytical and problem-solving skills.

Excellent communication skills in English.

Ability to work autonomously in a fast-paced, research-driven startup environment.

Important Calibration Points

This role is not designed for profiles mainly focused on fine-tuning, evaluation, annotation, or downstream applications of already-trained models.

Profiles whose experience is mostly centered on using existing models, without having designed and trained models from scratch, will not be aligned.

NLP profiles can be relevant, but only if they have strong technical depth in model architecture, deep learning, Transformers, and training. NLP backgrounds mainly focused on linguistics are not a fit.

The company is not looking for a profile mainly oriented toward software engineering. Strong engineering skills are valuable, but the core of the role is machine learning research.

Nice-to-Have Skills

Experience designing and running large-scale ML experiments using tools such as SLURM, PyTorch, DeepSpeed, or equivalent systems.

Experience with embedding models, representation learning, transfer learning, or foundation models.

Experience with structured data, tabular data, multimodal learning, vision models, NLP models, molecular synthesis, or other domains involving advanced deep learning research.

Open-source contributions, data science competitions, or strong public research output.

Track record of translating research into business or product impact.

Experience with Python and software development best practices.

Experience with C/C++ is a plus.

Familiarity with data formats and systems such as Parquet, SQL, NoSQL databases, and large-scale dataset management.

Ideal Background

The ideal candidate has completed a PhD in AI / Machine Learning and has then spent several years in a strong research environment, such as a top AI lab, deeptech company, academic research group, or industrial research lab.

A perfect profile would combine:

Strong ML research background.

Deep expertise in Transformers or foundation models.

Hands-on experience training models from scratch.

Exposure to pretraining.

Top-tier publications.

Ability to run rigorous experiments at scale.

Interest in applying research to a rare and high-impact data problem.

Location

Permanent position.

The role is open in Paris or London.

Hybrid work is preferred, especially for candidates based near the team.

Full remote may be considered for exceptional profiles based outside France or the UK.

Compensation

Base salary up to €100,000, depending on experience and seniority.

Equity may be included as part of the overall compensation package.

Benefits

Competitive salary.

Equity package.

Health insurance.

Meal vouchers.

French-level paid leave and time off.

Dynamic and ambitious research environment.

Flexible remote work arrangements depending on profile and location.

Recruitment Process

Fit interview - 30 minutes with the Chief of Staff.

Research fit interview - 45 minutes with the VP Research.

Debugging / coding test - 30 minutes.

Technical interview - take-home research topic followed by a 1h30 restitution with the research team.

Onsite step - lunch and interviews with the founders.

Reference calls.

J'ai volontairement rendu le poste plus sélectif que générique AI Scientist, pour éviter d'attirer des profils trop applicatifs, trop software engineering ou trop fine-tuning.

Profil recherché

The ideal candidate is a research-driven Machine Learning Scientist with a strong background in Deep Learning, Transformers, and foundation models. They should have a PhD in Machine Learning, Computer Science, AI, or a closely related field, with proven experience designing model architectures and training models from scratch, ideally including pretraining.

The role is best suited for someone who has published in top-tier ML conferences or journals and has worked in a strong research environment, such as an AI lab, deeptech company, academic research group, or industrial research lab.

The candidate should be hands-on with modern ML frameworks such as PyTorch or JAX, comfortable running large-scale experiments, and able to reason deeply about model performance, scalability, reproducibility, and evaluation.

This is not a fit for profiles mainly focused on fine-tuning, evaluation, annotation, downstream applications of existing models, or software engineering. NLP profiles can be relevant, but only if they have strong technical depth in model architecture, Transformers, and model training, rather than a primarily linguistic research background.

The strongest profiles will combine scientific excellence, hands-on experimentation, top-tier publications, and the ability to contribute to a highly ambitious foundation model research roadmap.