How can AI be used to extract structured data from commercial leases into a usable schema?

A lease scraper was built as an early AI learning exercise: to make lease data usable, it first had to be scraped into a defined data schema and stored in a table. The team defined the specific fields they wanted extracted and built a system prompt designed to be 'very skeptical' — the tool attempts to match against 16 defined fields but only populates a field if it has high confidence in the extracted value, otherwise it leaves it blank rather than guessing.