MLOps Engineer

We are seeking a talented MLOps Engineer to take full ownership of the AI pipelines that power our computer vision models. You’ll design and operate a containerized training/inference stack that runs both on a local GPU workstation cluster (multiple workstations with multiple GPUs each) and in Google Cloud. Your mission is to streamline the entire model lifecycle - from data ingestion and feature build, through training, evaluation, packaging, deployment, and monitoring - so researchers and engineers can iterate quickly and ship reliable models to production.

You will build robust orchestration and observability around our pipelines, implement resource-aware scheduling for heterogeneous queues, and lead the rollout of model/experiment tracking and performance analytics. You’ll also own the evolution of our documentation to ensure the platform is easy to understand, extend, and support.

• Hands-on MLOps experience building and running ML pipelines at scale (preferably computer vision) across on-prem GPUs and a public cloud (GCP preferred).
• Strong with Docker and Docker Compose in local and cloud environments; solid understanding of image build optimization and artifact caching.
• GitLab CI/CD expertise (modular templates, YAML optimization, build/test stages for ML, environment promotion).
• Proficiency with Python and Bash for pipeline tooling, glue code, and automation; Terraform for infra-as-code (GCP resources, IAM, networking, storage).
• Experience with orchestration: one or more of Airflow, Argo Workflows, Kubeflow, Ray, or Prefect.
• Experience operating GPU workloads: NVIDIA driver/CUDA stack, container runtimes, device plugins (k8s), multi-GPU training, utilization tuning.
• Observability & monitoring for ML and infra: Prometheus/Grafana, OpenTelemetry/Loki (or similar) for metrics, logs, traces; alerting and SLOs.
• Experiment tracking / model registry with tools like MLflow or Weights & Biases (runs, params, artifacts, metrics, registry/promotion).
• Data versioning & validation: DVC/lakeFS (or similar), Great Expectations/whylogs, schema checks, drift detection.
• Cloud services: GCP (Compute Engine, GKE or Autopilot, Cloud Run, Artifact Registry, Cloud Storage, Pub/Sub). Equivalent AWS/Azure experience is acceptable.
• Security & compliance for ML stacks: secrets management, SBOM/image scanning, least-privilege IAM, network policies, artifact signing.
• Solid understanding of containerized deployment patterns (blue-green/canary), rollout strategies, and rollback safety.

We believe great work starts with feeling valued and supported. That’s why we are building an thoughtful, competitive benefits and perks to help you thrive - professionally and personally - through every step of your Career with us. You will be eligible for:

• Salary from 2,500 EUR to 5,500 EUR per month (before Taxes).
• A Birthday Gift.

After Probationary Period

• Health Insurance
• Health Recovery Days (which can be taken as you need)
• Paid Study Leave
• Funding for the purchase of Vision Glasses after one (1) year of service

Join us in Building a Cleaner, Smarter Future - one quality process improvement at a time.

• Own the end-to-end ML pipeline for computer vision: data prep, training, evaluation, model packaging, artifact/version management, deployment, and monitoring (local GPU cluster + GCP).
• Design and maintain containerized workflows for multi-GPU training and distributed workloads (e.g., PyTorch DDP, Ray, or similar).
• Build and operate orchestration (e.g., Airflow/Argo/Kubeflow/Ray Jobs) for scheduled and on-demand pipelines across on-prem and cloud.
• Implement and tune resource allocation strategies based on current and upcoming task queues (GPU/CPU/memory-aware scheduling; preemption/priority; autoscaling).
• Introduce and integrate monitoring/telemetry for:
• job health and failure analysis (retry, backoff, alerts),
• data/feature drift and model performance (precision/recall, latency, throughput),
• infra metrics (GPU utilization, memory, I/O, cost).

Harden GCP environments (permissions, networks, registries, storage) and optimize for reliability, performance, and cost (spot/managed instance groups, autoscaling). Establish model governance: experiment tracking, model registry, promotion gates, rollbacks, and audit trails. Standardize CI/CD for ML (data/feature pipelines, model builds, tests, and canary/blue-green rollouts). Collaborate with CV researchers/engineers to productionize new models and improve training throughput & inference SLAs. Continuously improve documentation: update existing pipeline docs and produce concise runbooks, diagrams, and “how-to” guides.

• Kubernetes & Helm in production; NVIDIA GPU Operator, node labeling/taints, and MIG partitioning.
• Ray/Dask for distributed training/inference and hyperparameter sweeps.
• Feature stores (e.g., Feast) and streaming features (Pub/Sub/Kafka).
• Inference serving frameworks: TorchServe, Triton Inference Server, FastAPI + Uvicorn/Gunicorn, or Vertex AI endpoints.
• Batch & real-time pipelines: Apache Beam/Dataflow, Spark, or Flink.
• Cost optimization playbook on GCP: preemptibles/spot, autoscaling policies, right-sizing, per-project budget alerts.
• Testing for ML: pytest fixtures for data/model tests, golden datasets, regression tests, property-based tests.
• Experience with service proxies (Traefik/Nginx), DNS management, certificate management, and SSL/TLS automation.
• Familiarity with Edge/embedded deployments for CV models a plus.