colyptus.

Nº 002Finance · Document reading

From a pile of invoices to a searchable archive

Year
2026
Services
Custom software
Tools
C# / .NET, React Native, Vision-language model
From a pile of invoices to a searchable archive

The problem

A mobile app and archive that reads invoices, delivery notes, weighbridge tickets and fuel receipts from a photo, pulls out their fields, and puts every result in front of a person to approve.

Each document type asks for different fields: one wants amount and VAT, one a plate number and the goods, one tare and net weight. All of it was keyed into accounting by hand, and a template-bound reader could not keep up with that variety.

In three acts

From a pile of invoices to a searchable archive

Scene 01ACT I — BEFORETake 1
A pile of paper.

A pile of paper.

Invoices, delivery notes, weighbridge tickets and fuel receipts were keyed in by hand.

Camera: Scattered frame

Scene 02ACT II — TURNTake 1
The machine reads, a person approves.

The machine reads, a person approves.

A model recognises each document and pulls out its fields; nothing is saved unapproved.

Camera: Top down

Scene 03ACT III — AFTERTake 1
A searchable archive.

A searchable archive.

Approved documents are listed by type and can be searched.

Camera: Close-up

Approach

1

The machine reads

A vision-language model recognises the document type and extracts its fields.

2

A person approves

Nothing is saved on its own; a result enters the archive only once it is corrected and approved.

3

A searchable archive

Approved documents are listed by type and can be searched.

What we built

  • Mobile scanning app
  • Field extraction by document type
  • Review and correction screen
  • Invoice archive and search

Tools

  • C# / .NET
  • React Native
  • Vision-language model

Next project →

Woonzero: a universe of players' own

Something waiting to open?

An idea, a half-built product, a brand that needs a second life. Let's start with a short note.