Code
Coding Skills
- Deep learning: Python, PyTorch (from-scratch implementation, training, and evaluation of diffusion-like and transformer-based architectures) and related ecosystem (Lightning, WandB, etc.) — see Generative algorithms for structure prediction below.
- Scientific computing: Julia — see DFTK contributions and Bilayer graphene package below. Other contributions on Github. Fortran90, C++ — prior experience.
- High performance computing: Slurm (used extensively on the "Jean Zay" national supercomputer).
Software
Generative algorithms for structure prediction
I built from scratch a graph-based deep learning framework in Python to benchmark generative models for molecular structure elucidation. The framework implements different sampling and guidance methods for diffusion-like models, using Transformer architectures. It was trained and evaluated on the "Jean Zay" national supercomputer. This is joint work with Jérémie Cabessa, Thibault Charpentier, Marie-Pierre Gaigeot, and Mihai-Cosmin Marinica.The code is currently private and will be released upon publication (2026). It is available on request.
DFTK contributions
I made contributions to the Plane-Wave Density Functional Theory (PW-DFT) package DFTK in Julia language. Those include an interface with Wannier90 for Wannierization (later used to integrate the wannierization package Wannier.jl fully written in Julia language, in DFTK) and the implementation of the modified-operator method.Bilayer graphene package
With Étienne Polack, I developed a Julia package to compute band diagrams for twisted bilayer graphene, implementing the Bistritzer-MacDonald model and the model derived in Cancès, Garrigue & Gontier, Phys. Rev. B 107, 2023. Built as an overlay to DFTK. To be released upon publication — see chapter 5 of my PhD manuscript for details.