Nicolas Karasiak
Senior Geospatial ML Engineer
10 years building geospatial ML systems, from PhD research to production platforms
Open-source software adopted in 600+ scientific publications
Research contributions cited 400+ times
EXPERIENCE
EarthDaily, Senior Geospatial Data Scientist / ML Engineer Toulouse · September 2022 to Present
- Designed and built a production ML platform for processing hundreds of millions of field parcels, collaborating with software engineers and data scientists
- Created a label engineering framework preserving original ground-truth data while supporting flexible label mappings for training and evaluation
- Simulated EarthDaily’s daily satellite constellation by combining data from many sources with different spatial, temporal and spectral resolutions
- Fine-tuned and analyzed embeddings from geospatial foundation models (Tessera, Galileo) for crop classification and yield estimation
- Built the official EarthDaily Python package and developed scalable STAC/Xarray-based geospatial data workflows
Pixstart, Geospatial Data Scientist Toulouse · September 2020 to September 2022
- Built a rule-based scoring engine encoding weighted expert spatial rules (proximity to roads, land use, sensitive sites)
- Used it to find and rank priority areas, then built map products
Dynafor / INRA, PhD, Tree Species Mapping Toulouse · October 2016 to August 2020
- Mapped forest species from Sentinel-2 and Formosat-2 time series using phenology, feature engineering, and supervised classifiers (Random Forest, SVM)
- Showed how spatial autocorrelation biases ML evaluation, and developed validation strategies to address it
- Published peer-reviewed research (JOSS) and released Museo ToolBox, an open-source library for spatial cross-validation, model training, and map prediction on rasters
Guiana Amazonian Park, Intern Data Scientist Cayenne · March 2016 to September 2016
- Developed an open-source image classification framework in Python, R, QGIS, and OTB, released as the QGIS plugin Dzetsaka
- Applied it to vegetation detection and land-cover monitoring across French Guiana using multi-temporal satellite imagery
PROJECTS & OPEN SOURCE
- EarthDaily Python Client: official package, now the company-wide standard
- QGIS Dzetsaka Plugin: 200k+ downloads
- QGIS MCP Plugin: 100+ GitHub stars
- Museo ToolBox: peer-reviewed Python library for geospatial ML (JOSS)
SKILLS
- Programming
- Python, R
- Geospatial
- QGIS, GDAL, STAC, GeoParquet, Zarr, Xarray
- Machine Learning
- Scikit-learn, PyTorch, Lightning, time series
- Distributed Computing
- Dask, AWS, Docker, Argo Workflows
- Software Engineering
- System design, data platforms, performance engineering, open-source, Linux (daily driver)
- Languages
- French (native), English (professional)