The geocoding is just looping thru a json file and doing a weighted match of address parts.

I think I could put this in a postgres and query via SQL and speed it up.

I also am considering writing the geocoder in golang or rust to speed it up. But, the network request might be the limiting factor, which may mean I need to figure out a way to not go over the public network. Both services are in containers, but they aren't privately networked.

There's some interesting technical problems. First off, the corp db has about 7.5 million records.

I built the geocoder myself to overcome issues with google and opencagedata geocoders.

Geocoding takes half to three quarters of a second, but is single threaded (throw too many simultaneous requests at it and they back up and time put). the network request and db insert of the geocoded data take 1-2+ seconds.