Most AI resume and presentation tools work like: AI -> pick a template -> fill in content.

But LLMs are actually pretty good at generating and modifying raw HTML/CSS. In theory, you could:

1. Keep the doc as real HTML (or React) 2. Let AI directly adjust layout and styles 3. Iterate with prompts like “more minimal” or “stronger hierarchy” 4. Export to PDF at the end

So why are we still constraining everything to rigid templates? Is this about PDF rendering, layout instability, user trust, or something deeper around determinism and control?

Curious if anyone here has tried building something like this and what broke.