Author here!

That's true, and I actually used that dump[1] and OCR to automate about 80% of this project. The manual data entry was only required for the 20% that couldn't be found automatically.

Since this was just to fill in the gaps, I figured there would be diminishing returns on re-implementing fuzzy search on the data, but adding shortcuts to my manual workflow took less time for a reasonable speedup.

[1] I also used the JSON dump of this data that the FDA provides here: https://open.fda.gov/apis/device/510k/download/