SqaaS

SqaaS / Our work / Document processing

Built for a client

Legal & compliance

Parse thousands of documents with AI

Contracts, legal filings and identity documents arrive as PDFs full of tables, stamps, diagrams and scans. We built a pipeline that reads each page according to what is on it: plain text goes to fast, low-cost models; diagrams, images and scanned pages go to vision models that keep what is drawn, not only what is written. The extracted data is then checked against the other documents, to flag anything that doesn't match or doesn't look valid.

  • The right model for each page: text, tables, diagrams or scans
  • Information inside diagrams and images gets extracted, not skipped
  • Documents cross-checked against each other to validate them (KYC)

How we built it

  • Each page gets the model it needs. Pages are read according to what is on them, so plain text never pays for a vision model and a scanned page never goes to a text-only one.
  • What is drawn counts as much as what is written. Diagrams, stamps and images are read by vision models, and their information is extracted along with the text.
  • Documents check each other. The data extracted from one document is compared with the rest of the file, so a mismatch or an invalid value is flagged instead of slipping through.

The client stays confidential: we describe the solution, not who it was built for.

One problem, solved in production

This is the kind of work we take on as Service as a Software: you bring one problem, and we solve it with AI and put it into production. If your platform needs a capability like this for its own customers, we build it into your product with Component Builder.

More of our work