AI · OCR · Productivity

Yomasete OCR — Paper forms in, the same tidy fields out.

In development
  • Web
  • API

Yomasete OCR reads Japanese business forms such as invoices, delivery notes, quotes and receipts, and turns them into data your systems can use. Set up a form once, and every copy of it comes out with the same fields.

Illustration: the fields of a paper invoice turned into JSON data
※ Illustration
field types, from dates and amounts to checkboxes and tables
7
sample page is enough for AI to draft a template
1
export formats: JSON, CSV and searchable PDF
3

Why

Why we build it

Many offices still type in invoices and delivery notes that arrive on paper or by fax, one sheet at a time. Plain OCR only returns lines of text, while a business system needs the right value in the right field.

Features

What it does

  • One template per form

    A template is a reference image with named zones, so every document of that form becomes JSON with the same keys.

  • Handwriting models

    Dedicated models read handwritten digits, amounts and phone numbers, and English letters.

  • Tables and receipts

    It reads tables with or without ruled lines row by row, and splits a page holding several receipts into separate ones.

  • Rule fixes and master lists

    Rules clean up Japanese-era dates, full-width digits and yen amounts, and values are matched against master lists you upload as CSV.

How it works

How it works

  1. Upload one sample page and AI drafts a template; adjust the zones and save it.

  2. Upload scans, PDFs or phone photos of the same form as a batch.

  3. Each page is straightened, matched to the template and read zone by zone.

  4. Check the flagged values, approve, and export JSON, CSV or a searchable PDF.

Roadmap

Now and next

Now

In development

In development and running as a test deployment, with a web app and an API. We are also building a tablet demo app where invoices are written by hand and read by the same models.

Next

  • Finish workflows that link reading, checks and sending results to other systems
  • Let each organisation train text models on its own documents, not only digit models
  • Test the handwriting demo app on real tablets with a pen

※ Plans, not promises.

Built with

Built with

  • PaddleOCR
  • OpenCV
  • Python
  • FastAPI
  • PostgreSQL
  • Next.js

Contact

Questions about Yomasete OCR?

Product questions, partnerships or press: send us a message.

※ Image for illustration