I'm a ML Engineer / Data Scientist at Metafora Biosystems, a biotechnology company based at the Cochin Hospital (Paris 14). I work on METAflow, a novel AI-powered tool for flow cytometry analysis.
What I do:
- Develop machine learning algorithms and data processing pipelines in Python for cytometry analysis
- Build and maintain production-ready REST APIs using Django framework
- Design and deploy ML models on Google Cloud Platform (GCP), following ISO 62304 standards for medical device software
- Collaborate directly with users to gather feedback and translate requirements into actionable development tickets
- Participate in Agile/Scrum workflows, including sprint planning and backlog management
Former PhD student in Machine Learning / Statistics at Université Paris-Saclay, affiliated with the Institut de Mathématiques d'Orsay and part of the Datashape team at INRIA, under the supervision of Gilles Blanchard and Marc Glisse.
- Libraries: NumPy, Pandas, Scikit-learn, PyTorch, Matplotlib
- MLOps: Git, Docker basics, REST API development
- Cloud: Google Cloud Platform (GCP) - Compute Engine, Cloud Storage
- Methodologies: Agile/Scrum, ISO 62304 (medical device software)
- Version Control: Git
- Documentation: LaTeX, Zotero, Markdown
- OS: Linux, Windows (WSL)
- AI tools: Claude, Copilot
- AWS Machine Learning Engineer certification
- Docker & Kubernetes for ML deployment
- Advanced MLOps patterns
My doctoral research focused on quantification learning applied to cytometric datasets, particularly in the context of Metafora's METAflow software.
Key contributions:
- Developed methods for automatic analysis of flow cytometry data using machine learning
- Leveraged Reproducing Kernel Hilbert Spaces (RKHS) to embed and store high-dimensional features
- Created transfer learning techniques to analyze new samples based on previously analyzed ones
- Built frameworks to estimate population proportions in new samples
"Label Shift Quantification with Robust Guarantees via Distribution Feature Matching"
with G. Blanchard and B.-E. Chérief-Abdellatif
- Best Student Paper Award - Research Track at ECML/PKDD 2023
- ArXiv preprint
- Conference proceedings
Abstract: We propose a unified framework based on distribution feature matching that recovers estimators from both classification-based and statistical mixture modeling approaches to quantification learning. We provide robust theoretical guarantees under label shift and investigate misspecification scenarios.
- Machine Learning: Kernel methods, transfer learning, statistical learning theory
- Label Shift & Quantification Learning: Distribution matching, robust estimation
- Kernel Mean Embedding: RKHS methods, feature representations
- Applications: Flow cytometry analysis, biomedical data processing
- MLOps: Model deployment, API development, production systems
- ECML/PKDD 2023 - Turin, Italy
Label Shift Quantification with Robust Guarantees via Distribution Feature Matching
🏆 Research Track – Best Student Paper Award
- Journées de Statistique - Société Française de Statistique, 2023
- DataShape Seminar - INRIA, 2023
- Workshop FAST-BIG - Efficient Statistical Testing for High-Dimensional Models
- Séminaire des doctorants - Institut de Mathématiques d'Orsay, 2023
Co-organizer of the Master's seminar in Statistics and Machine Learning at Université Paris-Saclay (2022-2024)
Teaching Fellow - IUT Sceaux (2022-2023)
Mathematics for Management - L1 B.U.T GEA, taught by Pr. Patrick Pamphile