Aurimas Narkevicius
Data Scientist @B Cube, Tu Dresden
Signup · Get unlimited contacts
WORK HISTORY
Data Scientist @B Cube, Tu Dresden
3D Volume Segmentation Pipeline: Engineered an end-to-end ML pipeline for semi-automatic segmentation of TB-scale 3D volumes of porous calcite. Overcame the absence of ground truth by designing novel texture filters, with dimensionality reduction of filter coefficients (PCA, UMAP), clustering (HDBSCAN, K-Means), and Random Forest classification — requiring minimal user supervision.• Scalability & Hardware Optimization: Implemented Zarr out-of-core storage and random subvolume sampling to reduce RAM requirements, combined with GPU-accelerated computations via JAX, enabling full segmentation of 8-billion-voxel px) datasets on standard computers in under 1 hour.• Volumetric Analysis GUI: Developed a Python desktop application with TraitsUI and scikit-image, for interactive 3D volumetric data analysis, enabling non-technical researchers to independently characterize structural properties with automated post-segmentation workflows.• Hyperspectral Synchrotron Data Analysis: Analyzing combined X-ray based hyperspectral datasets × × arrays) to uncover spatial correlations between crystallographic structure and elemental composition, extracting physical meaning from complex multi-modal maps. Built parallel processing workflows (Joblib) to reduce iterative recalculation from hours to minutes.
EDUCATION
University of Cambridge
Bachelor’s Degree, Natural Sciences
University of Cambridge
Master’s Degree, Natural Sciences
University of Cambridge, UK
M.Sci & B.A., Natural Sciences
University of Cambridge
Doctor of Philosophy (Ph.D.), Materials Chemistry
ABOUT AURIMAS NARKEVICIUS
After years of research in materials science and soft matter physics at Cambridge and Dresden, I am now channelling my scientific rigour into machine learning, AI engineering, and ultimately robotics.One year ago I began this pivot — and it has been nothing but exciting and energising, though I have to admit that it has also been scary. Nevertheless, I now build ML pipelines for preprocessing and extracting information from 3D volumetric imaging data or synchrotron-based hyperspectral imaging, combining my deep understanding of physical systems with modern AI tools.My trajectory is clear: in five years, I want to be building the intelligence layer that turns raw sensor data — vision, touch, sound, force — into real-time robot decisions. I want to be the person who bridges software intelligence and physical hardware. I want to work with teams that deeply understand not only the engineering but also the world around it — and most importantly, respect the people in it.
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.