Physics-grounded AI for chemistry
Navigating chemical space
Empowering the chemical, pharmaceutical, and materials industries
Biomolecules
From small molecules to protein binding
Binding energies, conformer ensembles, and reactivity at quantum fidelity, fast enough for the throughput drug discovery actually runs at.
Molecular crystals
The same drug, as a solid
How a drug packs into a crystal decides whether it dissolves, stores, and lasts on the shelf. Competing packings can sit less than a kcal/mol apart, so separating them is a first-principles problem, and you have to solve it at screening scale.
Energy materials
Batteries, before the lab builds them
Screen electrolytes, interfaces, and cathode chemistries in silico at first-principles accuracy, and cut years of trial and error out of the loop. The same models that read the molecules read the solid too, from catalysts to semiconductors and ceramics.
01The Problem
Discovery is stuck between accurate and slow
Quantum mechanics predicts chemistry from first principles. The catch is the price of that prediction: even density functional theory (DFT) buckles at the scale real discovery needs. There are more than 10⁶⁰ drug-like molecules alone, so brute force was never an option. Promising candidates go untested, and progress waits on the next result from the lab.
Drug development
Lead optimization needs QM-grade energies for millions of candidates, far more than a DFT pipeline can work through.
Batteries
Electrolytes and interfaces call for long simulations of large systems. Ab initio methods run out at picoseconds and nanometres.
Solar & semiconductors
Screening photoactive materials and dopants in the lab alone costs years per device generation.
Advanced materials
From ceramics to polymers, structure and properties are linked across scales that no single method covers.
02The Solution
We built the first AI grounded in quantum mechanics
Our models learn from high-level quantum-mechanical data across chemical space, so they inherit the accuracy of first-principles methods at a fraction of the cost. Think of it as a digital twin of chemistry: you query it in seconds instead of queuing an overnight job on a DFT cluster.
Trained on beyond-DFT reference data
High-level quantum-chemical labels where they matter, solid baselines everywhere else.
Transferable across chemical space
One family of models spans organic molecules, biomolecules, and inorganic solids, with no refitting for each new system.
The physics is in the architecture
Symmetries and conservation laws are baked into how the model is built, so it stays reliable on molecules it has never seen.
10⁶⁰
molecules in drug-like chemical space
~1 kcal/mol
chemical accuracy, the bar we build for
20+ years
of research in quantum chemistry & AI
03Technology
Physics-grounded machine learning
We build quantum-mechanical laws directly into our models. Statistics on its own can only recall what it has already seen; physics is what lets a model reason about what it hasn’t.
Faster R&D
Screen orders of magnitude more candidates a day than a DFT pipeline can handle.
Lower cost
A calculation that used to tie up a compute cluster now runs as inference in seconds.
Learns as it goes
Every simulation sharpens the models, and they stretch to new chemistries without starting from scratch.
Why does AI need physics?
Unlike language, chemistry can’t be scraped off the internet. Experimental data is scarce and expensive to produce, so a purely statistical model runs out of road quickly: it interpolates yesterday’s measurements, but it can’t reach the molecules nobody has made yet.
Build in physical principles like symmetries, conservation laws, and quantum mechanics, and a model can generalize from very little data. That is what keeps predictions accurate outside the training set, which is exactly where discovery happens.
Chemical space has no map, but it does have a law. Every one of its 10⁶⁰ molecules answers to the same Hamiltonian, and the variational principle is the compass: it says which arrangements of atoms are favourable, and how far any estimate can sit from the truth. A model that learns that physics, rather than the handful of measurements it happened to produce, can navigate the space instead of retracing it.
04The Team
Science, AI & engineering experts
Founded by researchers who helped invent machine learning in chemical space, an approach the field has since taken up broadly. We put science first, and we go after the hardest problems in chemistry, materials, and pharma.
Fig. 01
Dr. Jan Gerit Brandenburg
CEO & Founder
Senior Director, Merck KGaA
Fig. 02
Prof. Dr. Alexandre Tkatchenko
Chief Scientist & Co-Founder
Professor of Physics, University of Luxembourg
Fig. 03
Prof. Dr. Anatole von Lilienfeld
Chief Scientist & Co-Founder
Professor, University of Toronto & Vector Institute
Fig. 04
Dr. Philipp Marquetand
CTO
Senior Engineering Manager, Canva
Fig. 05
Dr. Grégory Fonseca
Senior ML Scientist
Fig. 06
Dr. Daniel Nagel
ML Scientist
Fig. 07
Dr. Stefan Heinen
Senior Computational Chemist
Fig. 08
Mirela Puleva, MSc
Research Scientist
108k
citations of the team’s published research
Source: Google Scholar · July 2026
05Get in touch
Navigate chemical space — together.
Tell us what you’re trying to discover, and we’ll show you what quantum-accurate simulation can do for your R&D.
Prefer to join the journey?We’re hiring