Content

I open sourced the code of this website. It is built with React and deployed on Github Pages. It might be a bit overengineered for a simple website without a backend, but it was a nice opportunity to get to know the Material UI package. Github Pages is free for open source projects. Feel free to fork the project and adapt it for your own purposes.

We built an advanced analytics platform helping hundreds of data professionals at the Dutch Police work secure and efficiently. The platform is based on Kubeflow, and integrated with all important police systems. My former colleague Steven Reitsma published an interesting (Dutch) blog about it. He also published a related article on one of our Istio use cases.

I wrote a whitepaper on machine learning model serving. It explains how model serving platforms can help organizations scale their machine learning capabilities efficiently. The whitepaper contains a conceptual description of what I think a model serving platform should look like. I deliberately kept out implementation details. These differ from company to company. Therefore, they are not useful to all readers. You can download the whitepaper for free from BigData Republic's website.

My colleague Bertjan Broeksema and I were invited to Demetrios Brinkmann's MLOps Community Podcast. We talked about the hurdles large organizations face when scaling their machine learning capabilities. We also discussed how these hurdles can be tackled. We had a great time. Not only could we share our vision on the topic with the MLOps community, but also the questions from the community helped us polish this vision. The podcast recording is available on various platforms.

Netflix open sourced their internal machine learning platform Metaflow. Due to my interest in MLOps, I wanted to dive into the tool right away. Thus, I wrote a review on it. Check it out on Medium.

At PyData Eindhoven 2019 I gave a workshop on scheduling machine learning pipelines using Apache Airflow. In the first plenary part of the workshop I explained what Airflow is and why I love the tool. In the second part, attendees worked on some exercises in their own Airflow cluster, which we deployed for them on AWS. My colleague Dick Abma also wrote a blog on this setup. The plenary session recording is available on Youtube.

During my research internship, we studied the Minimum Error Correction problem. This problem has been intensively studied in the computational biology literature and is also known in the clustering literature: essentially we are required to find two cluster centres such that the sum of distances to the nearest centre, is minimized. The paper contains proofs about the computational complexity of a variant of the Minimum Error Correction problem. It was published in COCOON, the journal of computing and combinatorics.

We built an order forecasting system for logistics service provider CB. We built a machine learning system that answered the question: "To what customers are we expected to deliver tomorrow and what quantity?" The system was developed on Microsoft Azure and yielded a 10% reduction in shipping costs.