ML Engineering
Building reliable computer vision pipelines from data preparation to evaluation and deployment.
Paris, France · Available September 2026
I build vision systems for difficult data.
Machine Learning & Computer Vision engineer working at the intersection of medical imaging, self-supervised learning and applied research.
SPATIAL / 3D / CLINICAL
EVIDENCE, NOT HYPE
01 / The work
From public benchmarks to private clinical cohorts, I turn complex visual data into reproducible experiments and usable models.
Ways of working
How I can contribute.
Applied Research
Designing experiments, benchmarks and reproducible methods for difficult visual data.
Technical Collaboration
Rapid prototyping, model evaluation and computer vision consulting.
Selected work
Research-minded.
Production-aware.
A selected record of research, internships and open-source work.
01
AP-HP · 2026
Pulmonary nodule segmentation
Guided 3D segmentation and downstream clinical modelling on a rare private cohort.
- Mean Dice
- 0.87
- HD95
- 1.58 mm
- Usable cases
- ~200
MedSAM + LoRA · nnU-Net baseline · DICOM/NIfTI harmonisation
02
Open source · Sequence modelling
Sign language recognition & translation
German Sign Language video to written German with MediaPipe landmarks and a Transformer.
- BLEU
- 27.4
- Parameters
- 27.1M
- Data subset
- 2,000
03
Open source · Medical imaging
SegProbe
Reproducible, model-agnostic prompt evaluation for medical segmentation.
A lightweight Python toolkit for comparing prompt protocols, measuring manual effort and stress-testing segmentation models across 3D image volumes.
View repository ↗
74%fewer manual boxes in a small public reproducibility run
04
LIRIS · Self-supervised learning
Handwritten text recognition
A self-supervised pretraining pipeline combining SimCLR and masked reconstruction with MAE / ViT, trained on historical Gallica / BnF documents and synthetic data.
~5%Character Error Rate after fine-tuning
05
Open source · MRI
Brain tumour segmentation
U-Net with a ResNet-34 encoder and knowledge distillation for medical image segmentation.
- Dice
- 0.915
- IoU
- 0.887
06
Academic project · Chest X-ray
Pathology detection under label uncertainty
Multi-label chest X-ray classification with EfficientNet / DenseNet and uncertainty-aware training.
- Mean AUC
- 0.86
- Pleural effusion
- 0.91
CheXpert · class imbalance · U-Zero / U-Ignore
02 / Profile
Applied research
with engineering
discipline.
I am completing an M2 in Vision & Machine Intelligence at Université Paris Cité. My work is shaped by hands-on experience at AP-HP and LIRIS, and by a conviction that ML systems should be measurable, transparent and useful.
Focus
Medical imaging
Computer vision
Self-supervised learning
Tools
Python · PyTorch
nnU-Net · MedSAM
SimpleITK · Git
Languages
French — C2
English — C1
03 / Contact