Orcrist Technologies - Geospatial Data Scientist

• Develop change-detection workflows, including Sentinel-2 time series, and distinguish meaningful change from seasonality, cloud and shadow effects, acquisition differences, and registration errors. • Build and evaluate object-detection, segmentation, and classification methods for satellite imagery, using statistical techniques, classical image processing, and machine learning where appropriate. Evaluate state-of-the-art deep learning approaches pragmatically against simpler baselines, weighing accuracy, label requirements, generalization, inference cost, and real production viability. • Assess the suitability of different sensors, resolutions, acquisition conditions, and processing levels for each use case. Extend methods across optical, multispectral, thermal, and SAR data as requirements develop. • Design preprocessing and feature extraction with data engineers: quality masking, compositing, co-registration, normalization, spectral indices, and sensor-specific corrections. • Combine raster outputs with vector and temporal data for spatial statistics, zonal analysis, anomaly detection, and comparison across areas and observation periods. • Build or source reference datasets and evaluation protocols. Use spatially and temporally separated validation, measure false positives and missed detections, and examine performance across regions and sensors. • Deliver traceable results with source references, timestamps, confidence or uncertainty measures, and documented limitations. Help analysts understand when a result needs closer review. • Package tested Python methods for repeatable batch processing or inference. Work with data and platform engineers on runtime, memory, monitoring, and integration into Sentinel workflows. • Formal university training in geoinformatics, remote sensing, Earth observation, applied mathematics, computer science, or a related discipline, combined with substantial hands-on geospatial or remote-sensing experience. • Strong Python and scientific computing skills, using tools such as NumPy, pandas, SciPy, xarray, GeoPandas, and Rasterio/GDAL. • Practical machine-learning experience with scikit-learn and PyTorch or an equivalent framework, plus TorchGeo or related geospatial deep learning packages, including training, evaluation, and adapting existing models. • A solid understanding of remote-sensing fundamentals: spatial, spectral, radiometric, and temporal resolution; coordinate systems; image alignment; and data-quality limitations. • Experience with satellite image analysis and at least one relevant task such as change detection, segmentation, object detection, or land-cover classification. • Sound statistical judgment around sampling, spatial autocorrelation, data leakage, class imbalance, uncertainty, and generalization to new places and acquisition conditions. • An engineering-minded approach to research: versioned code and data, reproducible experiments, tests, and clear explanations of how a method performs and fails. • Clear communication in English with both technical colleagues and domain specialists. Eligible to work in Germany. • Nice‑to‑haves • Experience working with thermal infrared imagery, processing and analysing SAR data, or combining observations from multiple sensors. Deep experience in one modality is valuable. • A PhD in geoinformatics, remote sensing, Earth observation, or a related field. • Experience with geospatial foundation models: fine-tuning models such as Prithvi or TerraMind, or using AlphaEarth Foundations embeddings for downstream analysis. Ability to evaluate whether these approaches improve on task-specific models with the available imagery and labels. • Experience scaling geospatial analysis with Dask, Apache Spark/Sedona, or Zarr. • Experience deploying and optimising GPU inference with PyTorch/CUDA, ONNX Runtime with TensorRT, or NVIDIA Triton Inference Server, including batching image tiles and managing GPU memory and throughput. • Spatial SQL with DuckDB or PostGIS, STAC-based data discovery, and exploratory work in QGIS or geemap; producing COG and GeoParquet outputs for downstream GIS use. • Experience working with commercial imagery from providers such as Airbus, Satellogic, SatVu, or ICEYE. • Experience working in defence and intelligence environments or on related projects. • Strong interest and practical ability in agentic software development: using coding agents to plan, implement, test, and review software, and keeping up with rapidly evolving tools, techniques, and trends.