Scientific data does not forgive sloppy modelling. The datasets are small and expensive, the effects are subtle, and the audience of reviewers, regulators or your own R&D lead will ask exactly the questions a demo never has to answer. Our methods come from this environment. They rely on models grounded in the physics and biology of the problem, uncertainty quantified on every prediction, and explanations an expert can interrogate. Our team spans applied biology to theoretical physics, and we publish in the fields we model.
Materials
Our materials work runs from production floors to R&D labs. It covers automated analysis of micrographs and sensor data, defect detection on the line, property prediction from composition and structure, and virtual screening across search spaces too large to measure, driven either by experimental data or by atomistic models from quantum-mechanical calculations. Our current flagship is a self-supervised foundation model for microstructure imagery. It is the model behind our one-day weld experiment and behind MicrostructureDB, described below.
Life sciences
We provide biostatistics for clinical and environmental studies, detection and classification of organisms, lesions and cells in biological imagery, and lab-workflow software in which language models draft the structured reports that experts sign. What this shares with everything else on this page is measured error rates and calibrated confidence. In this domain "the model said so" convinces no one, and it should not.
Products and projects
The fastest way to see what our materials AI does is to use it. MicrostructureDB is the productised form of the same foundation models we deploy in consulting, approaching public release.
MicrostructureDB
Microstructure analysis as a platform: upload micrographs, search by visual similarity, run foundation-model analysis without training anything yourself.
AI4MI course
Our ESF-funded training programme: AI for materials scientists, from working knowledge to working models.
AID4GREENEST
A Horizon Europe project on AI-accelerated characterisation of green steels, where our modelling is delivered at consortium scale.
Research alongside industry
We work on both sides of the applied-research boundary: consortium partner and subcontractor in European and Flemish projects (Horizon Europe, VLAIO, ESF), standing collaborations with the universities our team came from, and proposal support for companies that want a funded route to de-risking their AI plans. If your problem still looks like a research project rather than a contract, that is a conversation we know how to have.
How we can help
Data & ML
The modelling practice behind these applications: vision, language and prediction with uncertainty.
The method
How foundation models, domain adaptation, fast models and conformal calibration keep scientific rigor intact in production.
Development
Gathering new data, or visualising it in an app or on the web? We build and deploy the software around it.
Model validation & deployment
An independent review of a model you already rely on, covering calibration, evaluation leakage and shortcut learning. It takes two to three weeks at a fixed price.