Why Kamali keeps real Office files

For weeks the plan has been elegant. Kamali reads an organisation's PowerPoint and Word files with my own renderer, turns them into HTML, and from then on everything (search, previews, generation, layouts) works on that one text. New documents are laid out in HTML from a small set of standard layouts and exported back to Office.

Fifteen layouts

The layout work was real. I read a year of real decks and grouped every page: dozens of families, then 15 generic layouts and 10 themes. I went through the groupings one by one on a review page. Picture zones, a brand kit, footers, a character budget for every text box from its size and font. More decks came in later, and the counts held up: most pages fit a handful of layouts.

Then I converted the real decks

This week I converted a year of actual decks to those standard layouts. The result wasn't theirs anymore. Spacing a little off everywhere, pictures cropped a little differently, a chart that lost its exact look. Every page would need a person to fix it, which is the opposite of the point.

The renderer is good, and it's measured every day. But "good enough to show you a preview" and "good enough to become the document the organisation sends" are different bars.

Back to real files

So documents stay real PowerPoint, Word and Excel files. A copied slide keeps its original shapes and is fitted to a layout without dropping words. Big decks get their pictures recompressed instead of converted: a large report shrinks to a tenth of its size and is still the same PPTX. The original's hash is kept, so its facts don't have to be read and paid for again.

The renderer keeps the jobs it's actually good at: reading files for search, drawing previews, measuring whether new text fits a box. That's still a lot. It just isn't the document.