Allan Koetter
I build the data foundation, the platform underneath it and the models on top — and I like owning all three. Twenty years of production data work, an MSc in geology, a Diploma in Data Science and Machine Learning, and a year as an ML engineer forecasting time series in production.
Three areas that overlap more than they differ: a model is only as good as the data under it, and only as useful as the platform that runs it.
Data platforms built from the ground up: containerised open-source systems for ingestion, storage and publication, automated ELT from many sources with Python and Dagster, and the databases underneath — SQL Server, PostgreSQL and PostGIS. Everything as code, reproducible, and rebuildable from Git.
The half of machine learning that decides whether a model ever leaves the notebook: tracked experiments, versioned models, orchestrated training and served inference — and automating the search over models and features, so the effort goes into the problem rather than the tuning loop. Grounded in a year running a production forecasting model and migrating it to Azure Machine Learning.
Machine learning that treats space, time and topology as first-class dimensions: earth observation and change detection, vessel trajectories, graphs, and groundwater in the subsurface. It is where the geology and the data work meet, and it has its own page: spatio-temporal.org →
An MSc in geology from the University of Copenhagen, then hydrogeology, groundwater modelling and contaminated-site work — first as a geologist, then as a GIS officer and environmental case officer in Danish local government. Every dataset had a coordinate and most had a timestamp.
A decade as a data analyst at the Capital Region of Denmark moved the weight from the geology to the data: SQL Server, ETL, spatial analysis, and prioritisation and forecasting tools that management used to decide where the money went. A Diploma in IT from DTU, specialising in data science and machine learning, followed in 2020, and after it a year as an ML engineer at PostNord forecasting parcel volumes in production.
Since then the work has been platform-shaped: data transformation on a big-data platform, and now leading a public agency's data capability — the architecture and build of its data and analytics platform. Building the layer underneath and using it are, to me, the same job.
The written project reports from the IT diploma, published in full — five reports from 2017 to 2019, in Danish.
Object-oriented programming, web technologies and big data, closing with a final project that tested an unsupervised-learning algorithm against a groundwater catchment. The through-line is spatial data: a raster library in 2017, remote sensing of gravel pits and NMFk in 2019.