An automated screening workflow built for a lending platform, handling the first stage of every loan application: reading the documents, scoring the applicant, and routing the decision, before a human ever has to open the file.
Reviewing loan applications by hand is slow, and it's easy to miss things when you're doing it visually: a name that doesn't quite match between an ID and a bank statement, income that doesn't add up against spending, a document that's simply missing. Nothing was scored consistently, and there was no clear record of why one application was approved and another wasn't.
The build itself had its own hurdle: pulling clean, structured data (name, ID number, ID type, date of birth, income) out of unstructured ID photos and bank statement files, reliably enough to score on. That's the piece the AI document analysis stage exists to solve.
An end-to-end screening workflow that picks up a new application the moment it's submitted, reads the applicant's ID and bank statement with AI, scores the application against income, spending, and document consistency, and routes it to one of three outcomes: approve, decline, or send to a human underwriter for a genuinely borderline case.
This isn't a black box that spits out yes or no. Every decision comes with a documented reason, whether that's a strong application sailing through or a flagged one waiting on human judgment. The system does the reading and the first-pass thinking; people only step in where a person actually needs to.