Applied ML — computer vision and federated learning.
Engineer from Mines Paris – PSL (Data/AI option), currently doing an MBA at the Collège des Ingénieurs. I like ML problems where the data is messy, distributed or scarce — and I always measure against a baseline.
Simple-VFL — my CEA research internship made simple: in vertical federated learning, everybody runs the same frozen vision model locally and uploads the features once. No pixels shared, no back-and-forth. It beat the published baseline on the reference satellite benchmark, and the gain holds across seeds. Tables, ablations and caveats are in the repo.
Student data challenges — cancer detection on histology slides (0.702 public ROC AUC against a 0.631 baseline, validated across hospitals), and a 1st-place finish detecting chimney emissions from 16,557 images with a label-efficient pipeline (day/night normalization, temporal windows, anomaly detection).
Cantateo — a live Django app serving daily liturgical psalms with synchronized audio tracks, running on my own Linux VPS. My playground for shipping real software: PostgreSQL, Redis, Celery.
Python · PyTorch · scikit-learn · DINOv2 · Django · PostgreSQL · Docker · Linux