Photo of Jan Brandejs

Jan Brandejs

Scientific computing | Tensor algebra & Quantum chemistry

About Me

I am an interdisciplinary computational scientist, I focus on designing scalable tensor engines, which serve as the mathematical foundation for both quantum simulations and modern AI/Large Language Model (LLM) architectures. Currently a postdoctoral researcher in computer science at IRIT (CNRS, Toulouse), where I scale a GPU-based tensor engine for structured sparsity to exascale using StarPU. Previously, at the CNRS Laboratoire de Chimie et Physique Quantiques, I architected a tensor toolchain for an Advanced ERC grant — automatic optimization of tensor expressions and distributed memory tensor contractions on GPUs with excellent weak scaling. I was recently awarded the Marie Curie Fellowship Seal of Excellence (92% score) for architecting a novel HPC tensor library utilizing theory of representations and graph theory.

Download Full CV

Core Expertise

  • Tensor Engines & Architecture: Architecting novel libraries for tensor operations, automatic expression optimization, and graph theory-based load balancing.
  • Quantum Chemistry & Physics: Coupled Cluster methods, Density Matrix Renormalization Group (DMRG), and Tree Tensor Network States.
  • HPC & GPU Computing: C++, OpenMP, MPI, SYCL, HIP/CUDA, Kokkos, StarPU, and Fortran on EuroHPC-class systems (LUMI, Adastra, Frontier, Summit, Karolina).
  • AI & Machine Learning: Deep Learning (TensorFlow), Transformers/LLM architectures.
  • Performance Optimization: Hardware-aware programming, GPU profiling (Omniperf, HPCToolkit).

Professional Experience

Postdoctoral Researcher in Computer Science

Institut de Recherche en Informatique de Toulouse (IRIT), CNRS | 05/2026 – Present

With Prof. Alfredo Buttari. Shared role with Maison de la Simulation (Saclay) under the NUMPEX initiative. Scaling a GPU-based tensor engine for structured sparsity to exascale using StarPU.

Postdoctoral Researcher in Computational Science

Laboratoire de Chimie et Physique Quantiques, CNRS, Toulouse | 04/2022 – 04/2026

With Prof. Trond Saue. Architected and developed a novel tensor toolchain for an Advanced ERC grant. Focus areas: automatic optimization of tensor expressions in the Coupled Cluster method, and excellent weak scaling for distributed memory tensor contractions on GPUs.

Research Fellow

Wigner Research Centre for Physics, Budapest, Hungary | 01/2021 – 12/2021

Implemented a high-performance Tree Tensor Network State method in C++ utilizing a combined fellowship and mobility grant.

HPC C++ Developer (PhD Research)

J. Heyrovsky Institute of Physical Chemistry, CAS, Prague | 10/2018 – 12/2020

One of 3 core developers of the MOLMPS package, achieving the first efficient MPI parallelization of the DMRG method for chemists, scalable to over 100 nodes.

Founding Software Developer (Part-time)

SignoSoft, Prague | 01/2015 – 12/2018

Architected a multiplatform PDF signing app and built a machine learning-based car insurance recommender for VIG inc. using Recurrent Neural Networks.

Selected Publications

  • Brandejs, J., Saue, T., Gomes, A., Visscher, L., Bientinesi, P. (2026). Report on the second Toulouse Tensor Workshop. arXiv:2602.05490 [cs.MS].
  • Fabbro, G., Brandejs, J., Saue, T. (2026). The nuclear electric quadrupole moment of 87Sr from highly accurate molecular relativistic calculations. J. Phys. Chem. A 130 (16), 3187–3196; arXiv.
  • Brandejs, J., Hornblad, N., Valeev, E. F., Heinecke, A., Hammond, J., Matthews, D., Bientinesi, P. (2026). Tensor Algebra Processing Primitives (TAPP): Towards a Standard for Tensor Operations. arXiv:2601.07827; reference implementation.
  • Sehlstedt, P., Brandejs, J., Bientinesi, P., Karlsson, L. (2026). The software landscape for the density matrix renormalization group. Computer Physics Communications 324, 110136; arXiv. Shortlisted in "good reviews" by Prof. T. Nishino.
  • Brandejs, J., Pototschnig, J., Saue, T. (2025). Generating coupled cluster code for modern distributed memory tensor software. JCTC 21, 15, 7320–7334; arXiv.
  • Fabbro, G., Brandejs, J., Saue, T. (2025). Highly accurate expectation values using high-order relativistic coupled cluster theory. J. Phys. Chem. A 129 (30), 6942–6958.
  • Visnak, J., Brandejs, J., Mate, M., Visscher, L., et al. (2024). DMRG-tailored coupled cluster method in the 4c-relativistic domain. JCTC 20 (20), 8862–8875.
  • Brabec, J., Brandejs, J., Kowalski, K., et al. (2021). Massively parallel quantum chemical density matrix renormalization group method. JCC 42, 534–544.

Teaching & Scientific Leadership