Machine Learning for Layout Recognition
Since GrandTotal's early days, there was this vision: import an existing invoice as a PDF and create a functional layout from it. Now, only 17 years later, an early version is available. New users see it during onboarding. Existing customers can select a PDF via File/Import in the Layouts section.
The Problem
Creating layouts manually takes time: positioning text fields, choosing fonts, adjusting spacing, configuring tables. If you already have a PDF – why shouldn't the software derive a layout from it?
Because PDFs are complicated. They don't store "here's an address", only "here's text at position X/Y". No structure, no semantics.
The Solution
Machine Learning recognizes patterns: Where's the address? Which text blocks belong together? Where's the line items table? The technology analyzes the PDF, tries to understand the structure, and creates a layout framework.
The result is, with luck, a finished layout or at least a starting point for further work.
Original: Apple Pages Template
Converted: GrandTotal Layout
What Works
Well: Simple invoices with clear structure, readable fonts, clean tables. PDFs from professional software.
Less well: Complex layouts, multi-column design, unusual fonts, scanned documents, PDFs with lots of graphics.
Important
This is an early-phase experiment. It works better sometimes, worse other times.
Feedback via the Support page helps improve the technology.
How to Test It
New Users
Automatically during onboarding
Existing Users
In Layouts section via File/Import select a PDF
Privacy
All models run locally on your Mac
Missing Fonts
Substituted with available fonts
Feedback
Submit example PDFs helps improve the technology