We are looking for an experienced We are looking for an experienced ML Engineer / MLOps Engineer to join a cloud-native project for a German customer.
The role is strongly engineering-focused and involves building production-grade ML infrastructure, working with GPU workloads, ML pipelines, LLMs and large-scale data processing.
📍 Location: Germany
🗣 German: B2+ - must-have
🗣 English: B1+
📅 Estimated start: September 30, 2026
What you'll be working on
- Build and orchestrate ML pipelines using Kubeflow Pipelines (KFP v2)
- Train ML models on GPUs and manage GPU resources within Kubernetes
- Fine-tune transformers and LLMs
- Track experiments and models using MLflow
- Build classical ML models with XGBoost and CatBoost
- Process large datasets using SQL Server and DuckDB
- Develop Python-based pipelines, integrations and tooling
- Maintain high engineering standards through testing, clean code and CI/CD with GitLab CI
- Work in a secure, zero-trust / secure-by-default environment with network policies and restrictive container permissions
Requirements
What we're looking for
- Hands-on experience with Kubeflow Pipelines, ideally KFP v2
- Experience training models on GPUs
- Practical experience with LLM / transformer fine-tuning
- Experience with MLflow
- Strong knowledge of XGBoost, CatBoost or similar boosting models
- Strong Python engineering skills
- Solid SQL experience and understanding of large-scale data processing
- Experience with CI/CD, clean code and automated testing
- Production-grade ML/MLOps experience beyond notebook-based experimentation
- Experience working in enterprise or regulated cloud-native environments
Nice to have
- Experience with LLM pre-training, beyond fine-tuning
- GPU orchestration in Kubernetes
- Experience with zero-trust environments, network policies and restrictive container rights
- Knowledge of DuckDB
- Experience with modern Python tooling such as uv
Previous healthcare or billing domain experience is not required, but you should be comfortable quickly getting up to speed with a new domain.
Originally posted on