Best strategy for structured data extraction

Hi Everyone!

I am a Data Scientist working on a real estate search portal in Italy (similar to Zillow). We are now implementing free search. For the demo, the only goal was to take user input and populate a JSON file which would then serve as the search query. For this, I have parallelised several calls to chatGPT, each trained to extract 1 single feature.
It seems to me that for most of the binary features regex would outperform any model. I tried to add a classifier head to an italian bert: dbmdz/bert-base-italian-cased, but these did not perform extremely well. All I’ve found on the internet are classifications concerning sentiment based text, not concerning features. Does anyone have any Idea of what would be the best way to convert a free text search to a JSON like object for a query on a database?