Documents in, structured data out.

Situation

In most companies the data that matters sits inside documents, and someone pulls it out by hand, page by page. Innomarks builds Uncovio to take over that step: upload a document, mark what matters on the page, get structured data back. The reading itself is AI work by vision models on Innomarks’ backend. The surface where people see their documents, mark areas, and shape the result is our work.

Constraint

The backend and its API already existed, and the product had two frontend generations behind it. The third had to bring together everything Uncovio had learned by then: marking on the document, building templates, managing organizations, the usage numbers. And it had to stay type-safe against an API that does not stand still, but keeps moving with the product.

What we did

1. 01

   We built the document viewer. Pages render in the browser, someone drags out an area, and the vision models analyze exactly that section. Detected text keeps its coordinates, so every extracted value stays connected to its place in the document.

2. 02

   We built the template builder. Field by field, users define the structure they want extracted, and the platform suggests templates to start from.

3. 03

   We set up the administration behind it: organizations and roles, importing new users straight from Excel, and usage numbers per person and month.

4. 04

   And we set the whole thing up afresh as a third frontend generation on Next.js. The API client is generated straight from the backend’s OpenAPI contract, so frontend and backend cannot quietly drift apart, even as both keep moving.

[ Screenshot — template builder with a suggested extraction schema ]

Where it stands

Uncovio is in active development. The product grows on the current generation, and we keep building it.

Facts