Applied Mathematics → Agentic AI
Mohsen Sadr
Applied mathematician building physics-based agentic AI that delivers solutions for the fusion energy and space industries.
About
I'm an applied mathematician and the founder of PAI Dynamics, building physics-based agentic AI that delivers solutions for the fusion energy and space industries.
Before founding the startup, I spent a decade building high-performance simulation and machine-learning software at MIT, the Paul Scherrer Institute, and the Swiss Plasma Center (EPFL) for national research programs.
News
2025 – 2026- Jul2026PAI Dynamics wins Venture Kick Stage 1
Awarded CHF 10,000 in Stage 1 of Venture Kick, Switzerland's leading start-up support programme, to help advance our physics-based agentic AI platform for fusion energy and space design.
- Jul2026
New preprint: ORB5X, built with Agentic AI
“ORB5X v1.0: a performance-portable global electromagnetic gyrokinetic PIC code in C++/Kokkos built using Agentic AI” is now on arXiv. A modern C++17/Kokkos rewrite of ORB5, developed at PAI Dynamics with PhysicsCode and delivered to the Swiss Plasma Center and Max Planck Institute.
- Jun2026
EuroHPC compute grant
ORB5X received its first EuroHPC Joint Undertaking allocation: 8,600 node-hours across five European systems — Leonardo BOOSTER, LUMI-G, MareNostrum5 ACC, Karolina GPU, and Discoverer.
- Feb2026PSI Founder Fellowship for Physics-based AI Platform
Awarded a PSI Founder Fellowship (up to CHF 150,000) to develop an AI-based platform that accelerates and reduces the cost of physical simulations for fusion energy and space technologies.
Selected work
03 shippedPhysicsCode
A physics-aware agentic coding assistant for simulation and engineering software. It pairs frontier LLMs with domain knowledge of numerical methods, solver architectures, and HPC codebases — so scientists and engineers can iterate on simulation code through natural-language prompts.
ORB5X
A performance-portable rewrite of the ORB5 gyrokinetic particle-in-cell code in modern C++17 with Kokkos — global, electromagnetic, multi-species, for first-principles simulation of turbulence, transport, and energetic-particle dynamics in toroidal plasmas. Portable from laptops to NVIDIA/AMD GPU supercomputers.
Career
2013 – presentExperience
- Founder2026 –
- Founder Fellow2026 –
- Research Affiliate2025 –
- Scientist2023–25
- Fellow2021–23
- Scientific Collaborator2020–21
Education
- PhD, Applied & Computational Mathematics2020
- MSc, Simulation Sciences2017
BSc, Mechanical Engineering2013
Grants & awards
- 8,600 hEuroHPC JU compute allocation for ORB5X, 2026
- CHF 10kVenture Kick Stage 1, PAI Dynamics, 2026
- CHF 100kPSI / QBIT Capital Founder Fellowship, 2026
- €1.2mEUROfusion / SNSF consortium project, 2021
- €100kWalter Benjamin scholarship, German Research Foundation (DFG), 2020
Publications
18 papers- 01
Sadr, Lanti, Mishchenko, Wang, Villard — “ORB5X v1.0: a performance-portable global electromagnetic gyrokinetic PIC code in C++/Kokkos built using Agentic AI”
- 02
Sadr, Mishchenko, Hayward-Schneider, Koenies, Bottino, Biancalani, Donnel, Lanti, Villard — “Linear and nonlinear excitation of TAE modes by external electromagnetic perturbations using ORB5”
- 03
Donnel, Cazabonne, Villard, Brunner, Coda, Decker, Murugappan, Sadr — “Quasilinear treatment of wave–particle interactions in the electron cyclotron range and its implementation in a gyrokinetic code”
- 04
Farazi, Sadr, Kang, Schiemann, Vorobiev, Scherer, Pitsch — “Resolved simulations of single char particle combustion in a laminar flow field”
- 11
Sadr, Hadjiconstantinou — “A variance-reduced direct Monte Carlo simulation method for solving the Boltzmann equation over a wide range of rarefaction”
- 12
Sadr, Hadjiconstantinou — “Variance reduced particle solution of the Fokker–Planck equation with application to rarefied gas and plasma dynamics”
- 13
Sadr, Tohme, Youcef-Toumi — “Data-driven discovery of PDEs via the adjoint method”
- 14
Tohme, Sadr, Youcef-Toumi, Hadjiconstantinou — “MESSY estimation: maximum-entropy based stochastic and symbolic density estimation”
- 15
Sadr, Torrilhon, Gorji — “Gaussian process regression for maximum entropy distribution”
Talks & teaching
2017 – 2026Talks
- TSVV10 MeetingEUROfusion, onlineJul 2026
- Particles, Flows & Maps for Sampling Complex DistributionsLausanne, Switzerland posterNov 2025
- 5th Mathematical & Scientific Machine LearningNaples, Italy posterAug 2025
- 30th Biennial Numerical Analysis ConferenceGlasgow, UK talkJun 2025
- Swiss Plasma CenterEPFL — invited seminarApr 2025
- CSD Scientific RetreatPaul Scherrer Institute — invited talkMar 2024
- Machine Learning Seminar SeriesPaul Scherrer Institute — invited talkApr 2024
- Symposium of Center for Computational Science & TechnologyMIT — invited talkMar 2023
- 4th Mathematical & Scientific Machine LearningProvidence, USA posterJun 2023
- 19th European Fusion Theory ConferenceVirtual posterOct 2021
- Swiss Plasma CenterEPFL — invited seminarApr 2020
- 9th Int'l Congress on Industrial and Applied MathematicsValencia, Spain posterJul 2019
- 10th Int'l Conference on Multiphase FlowRio de Janeiro, Brazil talkMay 2019
- Mathematics Institute of Computational Science and Engineering (MATHICSE)EPFL — invited seminarJul 2019
- 3rd European Conference on Non-Equilibrium Gas FlowsStrasbourg, France talkFeb 2018
- Institute of Fluid DynamicsETH Zurich — invited seminarMay 2018
- Institute of Fluid DynamicsETH Zurich — invited seminarAug 2017
Teaching
Introduction to Computational PhysicsETH Zurich — Monte Carlo methods · slides'23–24- Computational Statistical PhysicsETH Zurich — rarefied gas & plasma · slides'24
- Computational Physics 1 & 2EPFL — advection-diffusion, chaotic systems'20–21
- Mathematical Foundations 1–5RWTH Aachen — numerical methods, linear algebra'17–19