Loom files store data matrices in single-cell RNA-seq and is what velocyto works with to do RNA velocity.

The loom file is here: http://pklab.med.harvard.edu/velocyto/hgForebrainGlut/hgForebrainGlut.loom

Basically, this file was made by different pipeline in a previous study (see the paper: RNA velocity of single cells, Nature) and the notebook basically leverages that loom file to figure out which barcodes, genes, etc. to use in the downstream analysis. That Nature study ended up with a 1720 barcodes × 32738 genes matrix -- and, for consistency, the notebook is filtering the data to use the same 1720 barcodes and 32738 genes.

For your own data, you'll ignore that stuff -- you do your own barcode/gene filtering rather than work off a pre-existing loom file.