Consider using fuzzy match scoring to protect against hallucinations

[ppinchuk]

This might be a simpler alternative to prevent hallucinations during text extraction than what we are using currently:

https://github.com/seatgeek/thefuzz