Contract intelligence
Scanned contracts become structured records. A person confirms. Then the deadlines run themselves.
The situation
A public body managing thousands of contracts, amendments, tenders and framework agreements as scanned PDFs, where a missed expiry or renewal window has legal and financial consequences. Nobody could read them fast enough to keep every date in view.
How it works
Read
OCR recovers text from scans and image PDFs; the model identifies the document type (contract, amendment, extension, tender, call, framework agreement) and tags it so downstream logic knows what it holds.
Extract
parties and signatories, legal form and tax ID, subject, amount and tax status, start, expiry and extension dates, registry and reference IDs, classification code. The registry ID links tender to contract to amendment; identity keys on tax ID, not name; expiry is a set of key dates, not one field.
Confirm and drive
the record waits at a human gate. Once confirmed it enters the database and feeds expiry and renewal alerts.
Outcome
90-95% extraction on scanned PDFs · seconds per document · zero unconfirmed records reach a deadline.
Hardest part
The documents fight back: scans, dense legal language, several document types that store the same fact differently, the same counterparty under more than one tax ID. Reading the text was not the hard part. Designing a schema that holds all of it without losing meaning was, especially dates, where a hard expiry, an "on completion" clause and a submission deadline cannot share one field. And because the output drives real deadlines, the model is not allowed to guess. The real work was knowing exactly where extraction stops being reliable and a person has to step in.
See how this maps to your business.
walk through this build · no obligation