Philipp Marquetand

Theoretical chemistry · Machine learning · Engineering leadership

I build AI that understands molecules – and AI that helps people design.

I am a theoretical chemist turned machine-learning engineer and engineering leader. My work sits where quantum physics, artificial intelligence and real products meet: models that predict how molecules and materials behave in a fraction of the time quantum chemistry needs, and generative AI that turns a simple idea into a finished design.

Part of my work is AI that splits a flat image into editable layers – background, objects and text – so every part can be changed on its own. Hover over the picture on the right to see it come apart.Tap the picture below to see it come apart.

Dr. Philipp Marquetand

What I work on

Molecules and materials at machine-learning speed

Quantum chemistry is accurate but slow. I develop machine-learning models trained on quantum-mechanical data that predict molecular and material properties orders of magnitude faster – so drug candidates, battery materials and solar-cell compounds can be explored at a scale that used to be out of reach.

  • Drug discovery
  • Batteries
  • Photovoltaics
  • Materials science

Generative AI for visual design

Large language models and diffusion models can suggest layouts, rewrite text, edit images and create new visuals from a short description. I work on making these systems fast and dependable, so anyone can go from an idea to a polished design in seconds.

  • Large language models
  • Diffusion models
  • Multimodal AI

From research to production

A model only matters once it runs reliably for real users. I lead engineering teams that turn research into products – from data pipelines and training to scalable GPU inference, monitoring and operations.

  • ML engineering
  • Inference at scale
  • Team leadership

Example · crystal structure prediction

From a molecule to a medicine

Whether a drug works as a tablet depends not only on the molecule, but on how its molecules pack into a crystal. Paracetamol, one of the most widely used painkillers, crystallises in more than one form – and the forms differ in how well they can be pressed into tablets. Machine-learning models trained on quantum-chemical data can rank such crystal packings in a fraction of the time conventional calculations need. That is one of the problems I work on today.

Paracetamol molecular crystal PARACETAMOL
C₈H₉NO₂
Paracetamol, C₈H₉NO₂ – herringbone packing (illustrative), one unit cell outlined.

Roots in quantum dynamics

Before moving to industry, I spent many years in academic research on how molecules respond to light and ultrashort laser pulses, and helped pioneer machine learning for excited-state molecular dynamics. That work resulted in around 100 peer-reviewed publications and open-source simulation software.

Research background →

Selected publications

  1. Machine learning spectroscopy to advance computation and analysis

    J. Westermayr, P. Marquetand

    Chem. Sci., 16, 21660–21676 (2025).

  2. Deep learning study of tyrosine reveals that roaming can lead to photodamage

    J. Westermayr, M. Gastegger, D. Vörös, L. Panzenboeck, F. Joerg, L. González, P. Marquetand*

    Nat. Chem., 14, 914-919 (2022). arXiv:2108.04373 [physics.chem-ph]

  3. Machine learning for electronically excited states of molecules

    J. Westermayr, P. Marquetand*

    Chem. Rev., 121, 9873-9926 (2021). arXiv:2007.05320 [physics.chem-ph]

  4. Machine Learning Molecular Dynamics for the Simulation of Infrared Spectra

    M. Gastegger, J. Behler, P. Marquetand*

    Chem. Sci., 8, 6924-6935 (2017). Supporting information arXiv:1705.05907 [physics.chem-ph]

  5. SHARC - ab initio molecular dynamics with surface hopping in the adiabatic representation including arbitrary couplings

    M. Richter, P. Marquetand*, J. González-Vázquez, I. Sola, L. González

    J. Chem. Theory Comput., 7, 1253 - 1258 (2011).

All publications →

Get in touch

For collaborations, talks or questions, write to me at .