2 ms·

Thanks. I was wondering if you're using a LLM to get unstructured data? Or how do you recognize that something is a location for example?

• Jabbs 2y ago
  Ah yea no LLMs but probably some of the similar ideas were used by having a validating dataset. I just have ~800 common locations (city, state, countries) that are parsed out of the text with validating logic wrapped around each one. Far from perfect but it is now pretty accurate. I found the index of the location matters a lot for accuracy. Its still a WIP