How can AI be used to extract structured data from lease documents?

The team built a lease scraper: you drop a new lease into the system, and it scrapes the document against a defined data schema of 16 fields. The system prompt makes the AI skeptical, meaning it only auto-populates a field if it has high confidence in the match. In a demo, it matched 14 of 16 fields (two, tenant contact email and phone, were manually removed before the demo). All scraped data goes through a human approval step before being saved. This process was described as an important learning exercise: to use lease data downstream, it first has to be scraped into a consistent data schema/table. Future development is planned around adding schema for tenant and landlord responsibilities, especially for commercial leases.