I'm Guillaume, a french research engineer specializing in Artificial Intelligence and Deep Learning.
In 2019, I graduated from ESILV Engineering School and obtained a Master 2 degree in Data Science from École Polytechnique.
Sensitive to the #AIforGood movement, I sought to apply my skills in the medical field.
I then embarked on a research internship where I compared various approaches to classify tumors using
MRI images and radiomic data. It was a good experience, but that remained too academic.
I wanted to join a project with the ambition to use these technologies to revolutionize current practices.
This is how I began my adventure at the young Parisian startup AZmed. Our goal was to assist doctors
(radiologists and emergency physicians) in analyzing their ever-increasing volume of medical imaging
while staffing levels remained stagnant. I joined a team of 6 people in a global context where no AI
had ever been deployed in a medical center. We were the first to do so in France, with the software
Rayvolve, which enables fracture detection on radiographs. For 4 years, I contributed to developing,
improving, and expanding the scope of analysis of the algorithms. Clinical studies showed a real gain in
time and precision when comparing a doctor alone to a doctor assisted by Rayvolve.
By the end of 2023, Rayvolve had grown to be used in routine clinical practice across over 500 medical centers
in 50 countries - a significant achievement for our small team. After this rewarding journey, I felt it
was the right time to pursue a long-held dream of mine: taking several months to travel and gain new perspectives.
Now back, I am ready for a new challenge that pushes the boundaries of what's possible with AI!
A powerful detection algorithm for one medical center, but unsuitable for another. This is one of the major problems faced by research teams wishing to deploy their models on a large scale. In this article, I introduce this challenge called Domain Adaptation in the context of medical imaging, based on my experience at AZmed.
Research report comparing two main approaches - Radiomics and Deep Learning - to detect malignant soft tissue tumors directly on magnetic resonance imaging.
Utilities to download (CLI) and load an MRI brain tumor dataset with Python, providing 2D slices, tumor masks and tumor classes.
Application of object detection / segmentation (MaskRCNN and RetinaNet) on Mars satellite images. The goal was to detect automatically craters.
Implementation of the reinforcement learning technique Q-Learning. The environment was MountainCar (OpenAI Gym), where the goal is to teach a car to swing from left to right so that it can climb the mountain.
- Python # Numpy, Pandas, Scikit-Learn, PyTorch, TensorFlow, Keras, Matplotlib
- Web development # HTML, CSS, JavaScript, jQuery, Node.js, Express.js
- Databases # SQL (MySQL, PostgreSQL), NoSQL (MongoDB), ElasticStack
- C# / Java
- R
- Git
- Jupyter
- LaTeX
- Unix
- RStudio
- VS Code
- Android Studio
- Adobe Photoshop
- 🇫🇷 French # native
- 🇬🇧 English # fluent
- 🇪🇸 Spanish # basic
Probably the most cited "interest" in people's resumes, and I'm no exception.
It may sound cliché but I really like being out there, discovering new places.
My type of travel is all about adventure.
I dedicated 2024 to traveling around the world. Key destinations included South America
(from Lima to Ushuaia without flying), French Polynesia, New Zealand, Japan, and Southeast Asia (see more here).
Previous adventures include a 1-month motorcycle tour of France and a 16-day trek in the Everest region.