Research Engineer, Machine Learning

About Mistral AI

Mistral AI is a pioneering company shaping the future of AI. They believe in the power of AI to simplify tasks, save time, and enhance learning and creativity.

Role Summary

The Research Engineering team at Mistral AI spans Platform (shared infra & clean code) and Embedded (inside research squads). Engineers can move along the research↔production spectrum as needs or interests evolve. As a Research Engineer – ML track, you’ll build and optimise the large-scale learning systems that power their open-weight models.

Responsibilities

• Accelerate researchers by taking on the heavy parts of large-scale ML pipelines and building robust tools.
• Interface cutting-edge research with production: integrate checkpoints, streamline evaluation, and expose APIs.
• Conduct experiments on the latest deep-learning techniques (sparsified 70 B + runs, distributed training on thousands of GPUs).
• Design, implement and benchmark ML algorithms; write clear, efficient code in Python.
• Deliver prototypes that become production-grade components for Le Chat and their enterprise API.

Requirements

• Master’s or PhD in Computer Science (or equivalent proven track record).
• 4 + years working on large-scale ML codebases.
• Hands-on with PyTorch, JAX or TensorFlow; comfortable with distributed training (DeepSpeed / FSDP / SLURM / K8s).
• Experience in deep learning, NLP or LLMs; bonus for CUDA or data-pipeline chops.
• Strong software-design instincts: testing, code review, CI/CD.
• Self-starter, low-ego, collaborative.

What we offer

• Competitive salary and equity.
• Healthcare: Medical/Dental/Vision covered for you and your family.
• Pension: 401K (6% matching)
• PTO: 18 days
• Transportation: Reimburse office parking charges, or $120/month for public transport
• Sport: $120/month reimbursement for gym membership
• Meal stipend: $400 monthly allowance for meals (solution might evolve as they grow bigger)
• Visa sponsorship
• Coaching: they offer BetterUp coaching on a voluntary basis