Samir
Darouich

Hi, I'm Samir, a PhD researcher at the University of Stuttgart working on machine learning for chemistry. I develop generative models, mostly based on flow matching and diffusion, that predict the three-dimensional structures of molecules, such as stable conformers and the transition states of reactions. Two directions I'm particularly excited about are one-shot methods, which generate a structure in a single step instead of many small ones, and physically inspired generative models, which learn from the potential energy surface itself rather than from atomic coordinates alone.

I came to machine learning from chemical engineering, where I started out simulating molecules and materials with molecular dynamics and developing force fields. During an internship and working-student position at BASF, I worked with Ratan Mishra on molecular modelling for cement chemistry and corrosion protection. My first real contact with machine learning came when I used it to parameterise force fields from scarce experimental data. Moving from engineering into AI meant learning a lot from scratch, and honestly, that's still the part I enjoy most.

My PhD is advised by Mathias Niepert (AI) and Johannes Kästner (theoretical chemistry). Transition states of chemical reactions are my main test case, since they are fleeting, rare in data and expensive to compute. During a research stay in Heather Kulik's group at MIT, I studied how well generative models generalise to chemistry outside their training data, work now published in Nature Computational Science. More recently, I've worked on one-shot generative modelling under symmetries, which will be presented at NeurIPS 2026, and on flow matching that is supervised with the curvature of the potential energy surface.

I expect to finish my PhD in June 2027 and I'm open to R&D roles in AI for science as well as postdoctoral positions. If you're working on related problems, I'd be glad to hear from you.

• Oct 2026ElemeNet: Multiscale Molecular Machine Learning with Uncertainty Quantification Across the Periodic Table, a collaboration with the Kulik group at MIT, was published in the Journal of Chemical Information and Modeling.
• Sep 2026New preprint: Curvature-Aware Flow Matching for Molecular Structure Generation, which teaches generative models the curvature of the potential energy surface.
• Sep 2026SymDrift, one-shot generative modelling under symmetries, was accepted at NeurIPS 2026.
• Aug 2026Robust generative transition-state models for unseen chemistry, the work from my MIT research stay, was published in Nature Computational Science.
• Feb 2026Adaptive transition-state refinement with learned equilibrium flows was published in the Journal of Chemical Information and Modeling.
• Sep 2025I started a three-month research stay in Heather Kulik's group at MIT.

1. 2024 – now

   PhD researcher, theoretical chemistry and machine learningUniversity of Stuttgart

   Machine learning for reactivity in computational chemistry, advised by Johannes Kästner (theoretical chemistry) and Mathias Niepert (artificial intelligence).

2. Sep – Nov 2025

   Visiting PhD studentMassachusetts Institute of Technology, Kulik group

   Studied how well generative transition-state models generalise to chemistry outside their training data.

3. 2023 – 2024

   Data stewardUniversity of Stuttgart, Collaborative Research Center 1333

   Built data models and workflows for FAIR data management in chemistry, and a Python module to set up and analyse molecular dynamics simulations.

4. 2022 – 2023

   Intern, then working studentBASF, Molecular Modeling & Drug Discovery

   With Ratan Mishra: developed a method to scan adsorption free energies of organic molecules on mineral surfaces, now part of BASF's modelling workflows, and studied how salt additives slow the reaction of tricalcium aluminate with water.

5. 2022

   Pro3 Research FellowUniversity of Stuttgart, Institute of Technical Thermodynamics

   Developed TAMie force field parameters for diols from quantum-mechanical conformer studies, using machine learning to cope with sparse experimental data.

6. 2018 – 2023

   BSc and MSc, Chemical and Biological EngineeringUniversity of Stuttgart

   MSc with grade 1.0. Thesis on vapour–liquid equilibria from molecular dynamics free-energy simulations, with Niels Hansen and Joachim Gross. Along the way: teaching assistant in thermodynamics and engineering mathematics, and working-student roles at Dürr and Bosch.

I'm always happy to talk about transition states, generative models for science, or possible collaborations. The easiest way to reach me is by email at samir.darouich@ki.uni-stuttgart.de, or you can find me on LinkedIn.