Scanning transmission electron microscopy (STEM) is used for spectroscopy, diffraction, and imaging of materials. It is possible to determine many structural properties of materials using 4-dimensional STEM images. "Bragg Disk" detection is an important part of this process and can require computationally expensive image-processing steps. 
Scientists at NCEM originally developed a serial version of the Bragg Disk detection algorithm using Python and NumPy, running as a Jupyter notebook, where electron microscope image data were rendered for inspection and analysis. The image processing could be easily vectorized and was a good candidate for parallelization with a tool like Dask. By enabling a parallel implementation for the Bragg Disk detection code through Jupyter at NERSC, we were able to significantly improve run times for large data:  processing a 300GB dataset went from days to a few minutes. This workflow was scaled up to run on 40 Cori nodes (1280 workers), with results collected and visualized in the Jupyter Notebook.