Research Engineer · Scientific Computing · Computational Mechanics · Image-based Modelling
Hi — I'm a research engineer working at the intersection of numerical simulation, computational geometry, and deep learning for scientific imaging. I currently build C++ / Python simulation software at CEA‑LIST, where I work on smooth‑surface ray tracing and Fermat‑path acoustic propagation for the CIVA simulation platform — applied to transcranial focused ultrasound therapy planning and industrial non‑destructive testing.
Over the last ~15 years I've built scientific software across four countries — finite‑element solvers from scratch, image‑based mesh generators, U‑Net pipelines for X‑ray µCT, hyper‑elastic material laws in Fortran, and most recently smooth B‑spline surface fitting with parallel ray tracing. Same engineering profile every time: build a smooth, accurate representation of a discrete geometry so that physics simulation on top of it gets the right answer.
Below is a short tour of who I am, what I'm working on now, and what I've shipped before.
Sole developer of the smooth‑surface fitting + Fermat‑path acoustic‑propagation module being integrated into the CIVA simulation SDK. The pipeline fits a Multilevel B‑spline (MBA) surface to CT‑segmented skull meshes, ray‑traces against it with two interchangeable backends (custom Newton; Intel Embree BVH + Newton polish), and solves the multi‑surface Fermat path for ultrasound propagation through patient‑specific anatomy.
Projects I've personally built and shipped. Most of the code is private (industry partners — Safran Aircraft Engines, Siemens PLM, CEA‑LIST), but vox2tet is open source on GitHub and several associated papers are freely available.
CEA‑LIST · 2026 — present
C++20 / Python pipeline that fits smooth MBA surfaces to CT‑segmented meshes, ray‑traces against them with two interchangeable backends, and solves multi‑surface Fermat‑path acoustic propagation for transcranial focused ultrasound and adjacent NDT applications.
C++20EigenEmbreeTBBPythonSciPytFUS
LMPS / ENS Paris‑Saclay · Safran Aircraft Engines
Fortran VUMAT material laws (hyper‑elastic LAMCOS extensions + a new Mohr‑Coulomb elastoplastic law co‑developed with PhD student Lucas Pitta) for explicit‑dynamics forming simulation of Safran fan‑blade‑class dry textile parts. Abaqus + LS‑DYNA HPC runs (40‑CPU shared memory), Python Gauss‑Newton macro‑scale parameter identification, Matlab CT‑image deformation analysis.
FortranVUMATAbaqusLS‑DYNAPythonHPC
Mines ParisTech CMM (PSL) · Safran Aircraft Engines
Sole author of YarnPath, a regression‑CNN pipeline (TensorFlow / Python) that reconstructs yarn centrelines from µCT scans of 3‑D woven textile composites used in LEAP‑1A / LEAP‑1B aircraft engine root blades (A320neo, 737 MAX). 2‑D / 2.5‑D / 3‑D U‑Net ablation, distance‑transform and heat‑flow target encodings, and a confidence‑estimation method that works on inputs without ground truth.
PythonTensorFlowU‑NetµCTCUDA
Published in Composites Part A 186 (2024) 108396.
2018 (Python) — 2026 (C++ rewrite) · MPL‑2.0
C++17 pipeline that converts labelled 3‑D voxel images (TIFF) into high‑quality conforming tetrahedral meshes for finite element simulation. Multi‑material marching cubes (Wu–Sullivan LUTs), Laplacian + tangential surface smoothing, iterative remeshing, sliver‑triangle repair, TetGen tetrahedralisation. The Python original (2018) underpinned the Composite Structures 2022 paper; the C++ rewrite (2026) is the long‑form successor.
C++17CMakeEigenTetGenlibtiffOpenMP
github.com/skenop/vox2tet · Composite Structures 298 (2022) 116003
Ghent University · UGCT · Siemens PLM
Three‑and‑a‑half‑year postdoc on TensorFlow‑based U‑Net µCT segmentation + image‑based meshing + image‑based FE modelling of carbon‑fibre 3‑D textile composites and short‑fibre‑reinforced polymers. Three first‑author peer‑reviewed papers (Materials 2020, Composite Structures 2021 & 2022) in collaboration with UGCT and Siemens Industry Software. Synthetic image generation for training‑data augmentation, eliminating the manual‑annotation bottleneck. HPC training on the VSC (Vlaams Supercomputer Centrum) Ghent cluster.
PythonTensorFlowU‑NetµCTHPCCUDA
Institut Pprime (CNRS UPR 3346) · LABEX
µCT‑image‑based multi‑physics FE modelling of moisture diffusion and hygro‑thermo‑mechanical response in 2‑D / 3‑D carbon‑fibre textile composites. Custom Matlab / Python image‑to‑mesh pipeline (the direct seed of vox2tet), Abaqus with Fortran user subroutines (UEXPAN, ORIENT, USDFLD), and full periodic‑BC homogenisation. Published in IJSS (2018) and Composite Structures (2019).
MatlabPythonFortranAbaqusCGAL
Karlsruhe Institute of Technology · Institute of Engineering Mechanics
Five‑year from‑scratch Matlab FEM codebase for composite‑ microstructure optimisation (volume fraction + fibre orientation, gradient‑based with analytical sensitivities), multi‑scale homogenisation (Mori–Tanaka, Reuss–Voigt, secant nonlinear, RVE), Kelvin‑foam modelling, and CT‑image‑based voxel FEM scaled to 15.6 M elements. Experimental validation by synchrotron X‑ray diffraction on squeeze‑cast Al‑Si12 / Al₂O₃ metal‑matrix composites. Collaborations with Ole Sigmund (DTU) and Aberdeen CEMINACS.
MatlabMathematicaAbaqusStressCheck
Lviv National University · IAPMM NASU
From‑scratch C++/MFC Windows desktop solver implementing P1 finite elements with Ruppert Delaunay meshing, a posteriori adaptive refinement, and a hand‑written half‑band Gauss solver for the steady 2‑D convection‑diffusion‑reaction PDE. Thesis defended in Lviv, 2008 (157 pp., 40 figures, 168 references).
C++MFCFEMAdaptive