Skip to content
ENRU
All workStudio concept · 2026
Documents / AI

Dossier: AI reads the invoices, a person signs off

You upload an invoice and get the fields pulled out with a confidence score for each. You check the doubtful ones and send the result to accounting.

Watch the demo
Before

The accounting team gets dozens of invoices a day and retypes the details by hand. One wrong digit can mean hours of emails to sort it out.

After

We designed a service where the document is uploaded, the fields are filled in automatically, and a person only checks the ones the model isn’t sure about.

From a scan to a line in the books

Studio concept · fictional data · no network or database

  1. The user picks a sample invoice, and it opens next to the form.

  2. The supplier, tax ID, amount and date are already filled in. Each field has a confidence score, and the doubtful ones are highlighted on the document.

  3. A person fixes the doubtful values and confirms. The data goes to accounting along with a record of who checked what.

More product screens

01 / 04

Panel · document queueEvery incoming invoice, UPD and act with its status and model confidence: people only get the ones where the model is unsure.

Decisions

  1. Confidence for every field

    Along with each value, the model says how far it can be trusted.

    Why we chose this

    The reviewer only spends attention where it’s needed.

  2. Highlights on the document

    Click a field and you see where on the page the value came from.

    Why we chose this

    Checking a value takes seconds, with no need to read the whole document.

  3. Review log

    It records who corrected what, and when.

    Why we chose this

    A person stays responsible for the result, and you can see and count the model’s mistakes.

Product UI kit

A light grey work surface, violet for the AI’s suggestions, and clear, easy-to-read numbers.

Light grey + violet

Under the hood

An LLM extracts the fields and the formats are validated. The document and its highlights sit on an SVG layer, and the export to accounting goes through an API.

Product layers
  1. LLM
  2. OCR
  3. SVG
  4. API
  5. Postgres

Similar task?

Next caseCedarOrder service for a furniture workshop

Your project could be next

Tell us what you need, and we’ll show you how it could look.

Message on Telegram
Where to reply
What you’re interested in · optional
Project estimate

Send us your project

Describe the job in your own words and attach whatever you have: a brief, mockups, spreadsheets or a link to your site. We’ll take a look and tell you roughly what it could cost and where it makes sense to start.

Files · optional

Drop files here or PDF, Word, Excel, PowerPoint, TXT, PNG, JPG, WEBP. Up to 5 files, 20 MB in total
Where to reply

@ittectorittector@gmail.com