About

Foundation models @Waiv

I am a senior machine learning scientist at Waiv, the AI precision testing company formerly known as Owkin Dx, where I am research lead on foundation models. Waiv builds clinical-grade, AI-powered tests for oncology — biomarker detection, outcome prediction and treatment-response assessment — on top of one of Europe’s largest multi-institutional pathology data networks.

My own work sits at the point where representation learning meets the clinic. I released Phaet and Mascaret, fine-tuned pathology foundation models that are substantially more robust to scanner and staining variability without giving up downstream performance, and I built Histoboard so the community can compare published benchmark results for these models in one place. Alongside that, I am a core contributor on our diagnostic products — RlapsRisk BC and the MSIntuit Suite — from model development through international clinical validation, and lead scientist on pharma partnerships around immunohistochemistry readouts for antibody-drug conjugates. In practice that spans much of computational pathology: multiple instance learning, calibration and robustness, and cell and tissue segmentation.

Computer vision @Owkin

Between 2022 and 2026 I was a senior machine learning scientist at Owkin, a biotech company using causal, multi-modal AI to bring better drugs and diagnostics to patients. I was part of the Medical Imaging group, working on the analysis and representation of histopathology data.

There I built the Phikon family — one of the first publicly released sets of pathology foundation models — starting with Phikon, then Phikon-v2, a ViT-L pre-trained on 450M tiles drawn from 60,000 public whole-slide images, which matches models trained on proprietary data. I later published H0-mini at MICCAI 2025, distilling a large teacher into a model small enough to deploy while keeping a state-of-the-art robustness/performance trade-off. I also worked on the development and validation of AI-guided diagnostic tools, and on biomarker discovery for pharma partners.

Past experience in computer vision and medical research

After graduating I spent two years at Lille University Hospital in the Include team, the hospital’s data warehouse, created in 2018 and authorized by the CNIL to reuse patient data for clinical and methodological research. I built up expertise in (bio)statistics and machine learning (e.g. clustering, survival analysis, time series analysis) through collaborations with researchers and physicians, and worked on decentralized machine learning. In parallel I began specializing in computational pathology, through a two-year collaboration with Dr. Florence Renaud on predicting molecular subtypes in esogastric adenocarcinomas.

I then joined the ENDOMIC team (Inserm, INRIA, Lille University) as a data scientist, continuing on computational pathology and adding immunofluorescence image analysis through a year-long collaboration with the immunology institute of Lille University Hospital (Pr. Sylvain Dubucquoi, Pr. Vincent Sobanski).