What we work on
The range is wide on purpose. Physics and applied mathematics, machine learning and artificial intelligence, and the software engineering that takes a model from a notebook to a service other people can call.
What holds it together is not the subject. It is a preference for deriving a result rather than fitting one, for stating plainly what a model assumes, and for being specific enough about what it predicts that someone could show it to be wrong.
Where it ends up
Most of it exists because something being built needed it: a forecast that has to run every day, a service that has to hold up under load, a number someone is going to act on. That constraint is useful, because it rules out results that only work on clean data.
Part of the work is EU-funded and carried out with universities and industry partners across Europe.
How we publish
Papers go to a journal where the work suits one and to a preprint server otherwise. Some is internal, published here rather than submitted elsewhere. Whichever it is, the entry links to the paper itself rather than to a summary of it.
Where a paper is worth explaining at length we write it up in plain language: what the argument is, what it assumes, and what it predicts, without the machinery needed to prove it. The code goes out too, and Flama, the framework we serve everything through, is open source.