Senior Machine Learning Engineer

Atos; Brussel Hoofdstad; Hybride; EUR 90 000 - 130 000; Plein temps

Avantages offerts par ce poste

Platform-as-a-product culture

Autonomy over projects

Résumé du poste

Atos is seeking a Senior MLOps Engineer to own and advance the data platform’s ML lifecycle. You will work on a production-grade MLOps stack using Databricks and MLflow, deliver end-to-end pipelines, and collaborate with domain teams to scale across the organization.

You will own architecture, governance, and automation, building reusable templates and tooling. The role requires hands-on engineering, strong Python, and deep Cloud / Infra experience in Azure and Terraform.

Qualifications

• 5+ years in MLOps, ML engineering or platform engineering with production models.
• Hands-on Databricks and MLflow expertise is essential.
• Strong Python engineering and API/tooling skills.

Responsabilités

• Own the MLOps capability end to end, including design, implementation and roadmap.
• Assess and refine the platform alignment for ML lifecycle, landing zones, governance and automation.
• Deliver an incrementally adoptable ML lifecycle for one domain, then scale to all domains.
• Implement lifecycle on Databricks and MLflow: tracking, registry, feature tables, packaging, serving and monitoring.
• Build CI/CD promotion across environments for reproducible, auditable model releases.
• Use off-the-shelf capabilities where possible and custom components where needed.
• Enable automated retraining, drift detection and Lakehouse Monitoring with alerting and rollback paths.
• Productionise batch and near real-time inference.
• Treat models as data products with contracts, SLOs, lineage and cost signals.
• Provide cost visibility for ML workloads and right-size compute.

Connaissances

Python engineering

MLOps

ML engineering

Outils

Databricks

MLflow

Unity Catalog

Model Serving

Lakehouse Monitoring

Terraform

Azure

Description du poste

Senior MLOps Engineer — Data Platform

Location: Brussels, hybrid

Level: Senior

Languages: English required; French a plus

About the Team

We're building a data mesh on Azure and Databricks: domain teams own their data as products, and the central platform team gives them the paved road. We have an MVP data platform live with pilot teams onboarded. ML is the next capability: today it's a handful of models running outside the data platform, and we need a production-grade MLOps platform that domains can adopt on their own for these existing and new use-cases.

The Role

You’ll take ownership of the MLOps capability — its design, its implementation and its roadmap — and build it on top of the existing platform. You start with understanding the current platform (MLOps setup, landing zones, Unity Catalog, governance, automation), assessing what can be reused, and refining an ML lifecycle that fits it rather than sits beside it. Then you build it, with the domain teams as your first users.

Databricks and MLflow are at the core, so deep hands-on expertise there — Unity Catalog model registry, Model Serving, feature engineering, Lakehouse Monitoring, Asset Bundles — is essential rather than one skill among many. It's a hands-on engineering role: you write the pipelines, templates and tooling, and you embed with domain teams to see where they get stuck.

We value demonstrated experience over familiarity with the concepts.

What You’ll Do

• Own the MLOps capability end to end — its design, implementation quality and roadmap — and be the person accountable for where it is going.
• Assess the existing platform and refine the MLOps architecture to fit it: how models, features and experiments map onto the landing zone, Unity Catalog and governance model already in place, and what platform gaps need closing before ML can run there.
• Understand and refine the target ML lifecycle design with the platform architect and domain teams, then deliver it incrementally — first a working path for one domain, then a paved road for all.
• Implement the full ML lifecycle on Databricks and MLflow — experiment tracking, Unity Catalog model registry, feature tables, packaging, Model Serving and monitoring, tailored for regulated environments.
• Build controlled promotion across dev, staging and production with CI/CD (Azure DevOps / GitHub Actions, Databricks Asset Bundles), so model releases are reproducible and auditable.
• Deliver using off-the-shelf capabilities where they fit and custom components where they don't, and own that judgement call.
• Build automated retraining, drift and skew detection with Lakehouse Monitoring or equivalent, and the alerting and rollback paths that make them trustworthy.
• Productionise batch and near real-time inference.
• Treat models as data products — owners, contracts, SLOs and lineage from source data through features to consumers, with health and cost signals feeding the platform-wide observability and governance views.
• Give domains cost visibility for ML workloads — spend attributed per model and domain, right-sized compute, scale-to-zero serving, and surfacing idle endpoints and abandoned experiments.
• Manage ML infrastructure as code with Terraform, following platform standards, and review domain teams' ML deliverables.

What Success Looks Like in the First Year

• Ownership established — you are recognised by the platform team and domains as the owner of the MLOps capability and its direction.
• Design agreed — an ML lifecycle architecture that fits the existing platform, reviewed and backed by the platform architect and domain stakeholders, within the first quarter.
• First models in production — at least one domain running monitored, cost-visible models in production via the paved road.
• Smooth onboarding — a second domain can take a model from experiment to a monitored production endpoint without platform intervention.
• Roadmap delivered — the priority MLOps features e.g. near real-time inference, automated retraining shipped and adopted.

What We’re Looking For

• 5+ years in MLOps, ML engineering or platform engineering, with models you've built the delivery path for and supported in production.
• Someone who can own the design, implementation and roadmap of an MLOps capability through a shared vision — assessing an existing platform, designing to fit it, aligning platform and domain teams behind the direction, and delivering incrementally.
• Deep, hands-on Databricks and MLflow expertise — essential. MLflow tracking, models and registry in Unity Catalog; Model Serving, feature engineering, Lakehouse Monitoring, Workflows, Asset Bundles, system tables. You should be able to walk through ML platforms you've designed and operated on Databricks.
• Proven delivery of end-to-end ML pipelines using both managed services and custom components, with controlled promotion across environments.
• Hands-on model monitoring you built and operated — drift, skew, performance — not just configured.
• Strong Python engineering — code others can maintain.
• Solid Azure and Terraform for ML infrastructure.
• Near real-time inference and streaming (Event Hubs, Kafka, Spark Structured Streaming).
• Working understanding of data mesh and building platform capabilities for autonomous domain teams.

Nice to have:

LLMOps (evaluation, prompt and version management, RAG, agent frameworks, token cost), AWS, Kubernetes for serving, ML monitoring tooling (Evidently, Arize, WhyLabs), cost management of ML workloads with quantified results.

What We Offer

• A platform-as-a-product culture where adoption by domain teams is the measure of success.
• A team that values working code and clear documentation over slide decks, and gives you real autonomy over what you build.