15 years ago I wrote a demo for Azure HPC that did weather forecasting. A guy had a contact at NOAA that gave us a mishmash of shell scripts and Unix exes stitc

15 years ago I wrote a demo for Azure HPC that did weather forecasting. A guy had a contact at NOAA that gave us a mishmash of shell scripts and Unix exes stitched together to produce forecasts on huge compute clusters and my client wanted them ported to Azure to show off elasticity of the platform. It turned out to be a humongous PowerShell script that wrapped around all that crazy stuff downloading geography, unpacking, getting NOAA hourly data from their ftp, prepping it, forecasting (most intense compute step) and then chewing up results into a Bing map pngs at various scales. Much cygwin'ing was involved. The steps were named and script could fast-forward to them so I can debug each independently. The main forecast step was called tasseography and I was quite proud of my cleverness. The demo stayed up for like 6 years, surviving the project sponsor departure. I'd check in every once in a while and see people running forecasts in India or Europe or New York. I never even knew who paid for the compute, it was some random subscription.