Zaccharie Ramzi
AI Scientist at Mistral AI
Paris, Ile-de-France
Summary
👤
Senior🎓
Top SchoolZaccharie Ramzi is an AI Scientist based in Paris with 11 years of experience bridging deep learning research and production-grade ML engineering. He has advanced work on optimization and implicit models during a CNRS postdoc, applied physics-aware neural networks for 3D MRI reconstruction in a PhD role, and more recently focused on neural interfaces and alignment at Meta and Mistral AI. A careful coder who cares about pep8 and test coverage, he contributes to prominent open-source projects such as google/jaxopt, improving differentiable optimizers and robustness of line-search routines. His background spans academia and startups, from deploying wearable-driven models for healthcare to high-performance training on HPCs, which gives him a rare fluency across theory, numerical optimization, and production constraints. Colleagues describe him as research-driven yet pragmatic, capable of turning complex mathematical ideas into well-tested, deployable code.
11 years of coding experience
6 years of employment as a software developer
Master 2 (M2) Mathématiques appliquées, Master 2 (M2) Mathématiques appliquées at ENS Paris-SaclayMachine Learning, Machine Learning at Télécom ParisMathématiques, Mathématiques at Lycée Blaise Pascal
French, English, German, Arabic
Stackoverflow
2,196reputation
139kreached
63answers
22questions
Github Skills (17)
python10
differentiable-programming10
jax10
optimization10
data-structure9
algorithm9
algorithms9
data-structures9
machine-learning8
deep-learning7
keras6
tensorflow-datasets6
tensorflow6
plot6
numpy6
Programming languages (10)
Github contributions (5)
google/jaxopt
Oct 2022 - Jan 2023
Hardware accelerated, batchable and differentiable optimizers in JAX.
Role in this project:
ML Engineer
ML EngineerContributions:37 reviews, 8 commits, 15 PRs in 3 months
Contributions summary:Zaccharie primarily contributed to the development and improvement of optimization algorithms within the JAXopt library. They added features like gamma to the L-BFGS state and introduced the ability to stop L-BFGS after a line search failure. Additionally, the user corrected examples for implicit differentiation and enabled the return of auxiliary values from linesearch methods, enhancing functionality across multiple optimization algorithms. Their contributions include bug fixes and adding non-regression tests, indicative of a focus on code quality.
pytorchdifferentiabledifferentiable-programmingautomatic-differentiationdeep-learning
zaccharieramzi/starter-academic
May 2021 - Oct 2023
Contributions:78 pushes, 1 branch in 2 years 5 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.