About us
A Nordic deep-tech ML company that treats uncertainty as a first-class citizen, not an afterthought.
The thesis
Most production ML today ships overconfident predictions. In healthcare, finance, and industrial settings, that's not just suboptimal, it's dangerous.
We founded Desupervised in 2018 on a simple thesis: serious production ML needs first-class, well-calibrated uncertainty on every prediction. Not as a post-hoc wrapper, as a property of the model.
That thesis took us from bespoke consulting on a handful of hard problems to running our own research programme and publishing an open-source probabilistic programming language.
Copenhagen, Denmark.
In Michael's own words
Two short explainers from Michael Green, our CEO, on the ideas behind what we build.
What is Bayesian AI?
Michael Green explains what Bayesian AI is, why uncertainty is the missing piece in most production ML, and how it changes the way models behave under real-world conditions.
The different levels of AI
A short tour of the different levels of AI: what people mean when they say "AI", where machine learning fits, and why the distinction matters when you decide what to deploy.
Borch, our open-source PPL
Borch is the public expression of our thesis: a deep universal probabilistic programming language built on PyTorch, with one-line conversion of any neural network into a Bayesian model. It's released under an open licence and documented in a peer-indexed preprint.
Everything we ship runs on Borch. That's how we guarantee the same uncertainty properties whether you're using Alviss AI, AIChemy, or a model we co-developed with your team.
Two productised verticals
On top of Borch we run two commercial products, each targeting a domain where uncertainty-aware modelling is a competitive advantage.
Alviss AI
Real-time probabilistic forecasting for commercial decisioning: marketing-mix modelling, demand planning, customer behaviour, portfolio volatility.
Visit alviss.ioAIChemy
Equivariant deep probabilistic graph neural networks for drug-discovery virtual screening, benchmarked against classical pipelines in peer-reviewed work.
Visit aichemy.ioUniversities and research clusters we work with
Our research programme runs in collaboration with leading Nordic universities and innovation clusters across Sweden and Denmark.
Ready for the next step?
Tell us the decision you need to make with confidence. We'll tell you honestly whether a model can help.