Matthijs Keep

Building ML systems

I build machine learning systems end-to-end: from data pipelines and model training to evaluation and production deployment. Strong focus on PyTorch, attention mechanisms, transformers, and the full MLOps lifecycle. I enjoy building systems that work at scale and thinking carefully about architecture.

Currently interested in: training efficient transformers, building production-ready systems, physics based neural nets, and experiment management. AI at ASML.

Core strengths

• End-to-end ML systems (data → model → deployment → monitoring)
• PyTorch
• Transformer architectures and attention mechanisms
• Training at scale (distributed training, mixed precision, gradient checkpointing)
• MLOps: orchestration, CI/CD, experiment tracking, model serving
• Evaluation methodology and benchmarking

Projects

End-to-end Kubernetes ML system

github repo

End-to-end, production-emulating ML system. Uses MNIST as example, but any dataset/model can be used with PyTorch. Built to run locally on Kubernetes on a MacBook (Apple Silicon). My first attempt at getting some experience with Kubernetes.