Scientific Software for Large-Scale Research Data
Research Software Engineer @ Institut Cochin · CNRS · France BioImaging (2024–2026)
MSc Bioinformatics @ Université Paris Cité
Paris, France

profile = {
    "role":    "Research Software Engineer · Scientific Computing · Distributed Systems",
    "focus":   ["Software Architecture", "GPU Compute", "Pipeline Orchestration", "FAIR Data Infrastructure"],
    "domains": ["Bioimaging", "Genomics", "Medical Imaging / ML"],
    "status":  "Open to Research Software Engineer roles · Paris / EU / international (2026+)",
}

I build software systems for large-scale scientific research, turning heavy, error-prone workflows into reliable, reproducible systems researchers can depend on.

My work sits at the intersection of software engineering and science, where I turn heavy research workflows into distributed, GPU-accelerated systems that run in production rather than as standalone prototypes. I mainly work with biomedical data (imaging, genomics, proteomics), where acquisition volumes grow faster than they can be processed.

My edge: I write production-grade code and understand the science behind the data.

OpenCID (Open Collaborative Image Database) is Institut Cochin's platform for storage, processing, visualization and FAIR sharing of large-scale biomedical research data, run by the IMAG'IC Core Facility (France BioImaging).
Registered as a research resource (DOI 10.57889/cid).

I developed and maintained its processing and orchestration layer: a GPU scheduler and job lifecycle that let multiple researchers run heavy compute concurrently on shared servers directly from the browser. Within it, I automated the analysis tools (Huygens, Cellpose, ilastik, QuPath) and designed a plug-in architecture so each integrates the same way and new ones are fast to add.
For two years, I focused on making large 3D microscopy volumes processable under tight memory and compute limits, and built a browser-based 3D viewer to explore and analyze them. Poster 2025

Languages

Deep Learning & Computer Vision

Scientific Imaging & Bioimage Analysis
Cellpose · StarDist · Ilastik · Huygens · QuPath · ImageJ/Fiji · Bio-Formats · DICOM · OME-TIFF · OME-Zarr

Software Engineering & Infrastructure

Selected engineering and research projects across scientific software, imaging and computational biology.

Fork of Vol-E (Allen Institute) extended to open large OME-Zarr volumes packaged as local .zip, entirely in the browser, with no server.
Client-side ZipStore (lazy chunk-level reads) · per-object feature analysis (scatter, gating, CSV) · time-synced 3D tracking overlay · contributor to the OME-NGFF spec (RFC-9)
TypeScript · React · OME-Zarr · WebGL · Three.js

Adapting a mammography foundation model (Mammo-CLIP, EfficientNet-B5) to screening mammography: multi-task heads, controlled ablations and probability calibration.
AUROC 0.897 (breast level) · leakage-free patient split · reproducible via pixi + Docker
PyTorch · efficientnet_pytorch · scikit-learn

Protein structure recognition by threading a sequence through known templates, via double dynamic programming.
Faithful 1998 algorithm + modern optimizations · DOPE statistical potentials · Z-score validation · GPU acceleration
Python · BioPython

Benchmarking first- and second-generation protein language model embeddings.
Python · PyTorch

Always happy to connect, exchange ideas, or just talk science and code.