Urban wind simulation tools often require STL files as their input. The source data at rest is often large STL files representing the surrounding context of a site being evaluated. In traditional workflows, data preparation of slicing out the relevant context is done by the user on local hardware or in a VM.

When we moved a wind simulation workflow to the cloud, the memory requirements of this pre-processing overwhelmed the standard serverless offerings (Lambda function). We had to fall back to using a VM to handle this conversion, which doesn't scale horizontally and often goes unused.

Inspired by GeoParquet's bounding box around row groups, I devised an index footer for standard STL files which would allow a caller to use range requests to pull the subsection of the (pre-sorted) triangles which intersect an area of interest, decreasing the disk and memory usage of these types of workflows.

I'm currently testing the reader and writers, and it will be publicly available by the time the conference begins