Comment by adrianparlow

Without getting into too much detail, our approach is more metadata-based as screenshots/OCR are pretty noisy and expensive to work with. Shoot me an email if you like, happy to chat! [email protected]