Thomson Reuters needed high-fidelity training data for a document layout segmentation model capable of interpreting visual structure and contextual meaning across diverse document types, since traditional tools could capture text but not how it was arranged or its relevance. Invisible built a flexible data annotation framework with an advanced schema to represent document components like tables and forms, with pixel-level accuracy, establishing a seamless workflow for expert annotation and review.
Thomson Reuters needed high-fidelity training data for a document layout segmentation model capable of interpreting visual structure and contextual meaning across diverse document types, such as invoices and forms. Traditional OCR (optical character recognition) tools could capture text, but not how that text was arranged and its relevance. This made understanding the contextual significance of tables, form fields, and complex questionnaire responses difficult.
Invisible built a flexible data annotation framework and designed an advanced schema to represent document components – including tables, forms, and nested fields – with pixel-level accuracy. The team established a seamless workflow for expert annotation and review within Invisible’s platform, ensuring every dataset was consistent, and ready for ingestion into Thomson Reuters model pipelines.
The company is building AI for customers who need outputs they can trust, verify, and defend in high-stakes professional settings. Thomson Reuters calls this as Fiduciary-Grade AI™: systems designed for professional environments where accuracy, transparency, and accountability are essential. For these customers, AI has to do more than generate plausible answers. It has to understand the underlying structure and meaning of complex professional documents.
