Senior Data Infra Engineer for Scalable AI Compute

Mistral.ai; Paris; Hybride; EUR 90 000 - 140 000; Plein temps

Résumé du poste

Mistral AI is building the Data Infrastructure team to power large-scale training and fine-tuning for world-class AI systems. You will architect, scale, and operate massive compute fleets and data stores, enabling secure, governed access for MLOps and research teams.

You will own end-to-end lifecycle aspects—from migrating away from legacy schedulers to production-grade pipelines, to on-call rotations for critical training workloads, in a hybrid or remote EU/UK setup with monthly hub visits.

Qualifications

• 4+ years of experience in Data Infrastructure, MLOps, or Infrastructure Engineering.
• Experience or strong interest in supporting foundational compute and storage platforms.
• Proficient in Python and solving the "brittle data lake" problem with modern, columnar storage standards.
• Well-versed in Kubernetes-native tooling and debugging large-scale distributed systems across multi-cluster environments.
• Ability to build scalable, reliable, and secure systems from the ground up.
• Comfort with ambiguity in a rapid-growth AI environment.

Responsabilités

• Build & Scale: operate massive distributed compute and storage systems.
• Global Orchestration: architect and maintain multi-cluster orchestration layers across regions.
• Design Future-Proof Storage: transition to modern storage formats for exabyte-scale datasets.
• Platform Engineering: develop internal training platform across Kubernetes and SLURM environments.
• Metadata & Lineage: implement systems for data/model visibility and lineage.
• Operational Excellence: manage cloud-native deployments for scalable, reliable data platforms.

Connaissances

Python

Kubernetes

MLOps

Multi-cluster orchestration

Outils

SLURM

Description du poste

Mistral AI is building the Data Infrastructure team to power large-scale training and fine-tuning for world-class AI systems. You will architect, scale, and operate massive compute fleets and data stores, enabling secure, governed access for MLOps and research teams.

You will own end-to-end lifecycle aspects—from migrating away from legacy schedulers to production-grade pipelines, to on-call rotations for critical training workloads, in a hybrid or remote EU/UK setup with monthly hub visits.