stealth
Agentic AI for complex workflows in the pharma industry. Name and details when it's time.
about
I'm an ML researcher and engineer, based in Paris. I spent four years at Owkin working on machine learning applied to biology — building AI products for drug discovery, and later on agentic systems for biomedical research.
I'm now building in stealth, applying agentic AI to automate complex workflows in the pharma industry.
More broadly, I'm drawn to the intersection of AI and science: biotechnology in particular, and what these tools mean for human health and our understanding of consciousness.
Outside of work: martial arts (boxing and BJJ), music production, and reading across comparative philosophy, geopolitics, and literature.
building
work
Owkin
Led one of the teams building Owkin's Discovery Engine, an ML platform turning large-scale genetic data into new therapeutic target hypotheses. The platform anchored multi-million-dollar pharma partnerships and compressed target identification from years to months across several oncology programs. Later worked on agentic systems for biomedical research.
UC Berkeley
Deep generative models for single-cell genomics and cell lineage tracing, hosted by Nir Yosef and Romain Lopez. Best Paper Award at the ICML 2021 Computational Biology Workshop.
Hyperlex
Legal-tech startup applying NLP to contract analysis. Built a handwriting detection pipeline and benchmarked early transformer models (GPT-1 & 2) on legal use cases.
A*STAR
Unsupervised anomaly detection for medical images in Singapore. State-of-the-art results on CIFAR-10/100 and two ophthalmic imaging datasets; two publications on retinal image analysis.
research
Selected publications. Full list on Google Scholar ↗.
Comprehensive benchmarking of batch integration methods for spatial transcriptomics using a large-scale cancer atlas
ICLR 2026 LMRL Workshop
Benchmarks 11 representation-learning methods on Owkin's MOSAIC cancer atlas. Introduces a new metric for robustness to domain shifts and generalization to unseen samples.
Joint probabilistic modeling of pseudobulk and single-cell transcriptomics enables accurate estimation of cell type composition
ICML 2025 Generative AI & Biology Workshop
MixupVI: a deep generative model with a mixup-based regularizer that learns a latent space with an additive property, enabling reference-free deconvolution of bulk RNA-seq samples.
Robust evaluation of deep learning-based representation methods for survival and gene essentiality prediction on bulk RNA-seq data
Nature Scientific Reports, 2024
Systematic benchmark of representation learning approaches on bulk RNA-seq for predicting patient survival and gene essentiality. Code: owkin/drl-evaluation ↗.
Reconstructing unobserved cellular states from paired single-cell lineage tracing and transcriptomics data
ICML 2021 Computational Biology Workshop Best Paper
TreeVAE: a deep generative model that uses lineage tracing trees as structural priors to infer ancestral cell states that were never directly observed.
Towards practical unsupervised anomaly detection on retinal images
MICCAI 2019 Workshop
Transfer-learning approach to unsupervised anomaly detection for retinal disease screening, requiring no labeled abnormal examples.
Unsupervised deep learning for automated fundus image quality assessment in retinopathy of prematurity
17th International Conference on Biomedical Engineering, Singapore
Automated quality assessment of retinal fundus images to support screening for retinopathy of prematurity in newborns.
talks
- From patient data to agentic systems for drug discovery · Dust Engineering Night #11 · link ↗
- Comment Owkin accélère la découverte de médicaments grâce à l'IA (french) · Datascientest Webinar · link ↗
- AI-driven multi-modal target identification: from bedside to bench · GA4GH 2022 · link ↗
- Reconstructing unobserved cellular states · ICML Computational Biology Workshop, contributed talk · link ↗
education
-
ENS Paris-Saclay · MSc MVA -
CentraleSupélec · MSc Applied Mathematics -
Université Paris-Dauphine · BSc Applied Mathematics (top 5% of cohort)