Show HN: Snapdeck – Generating presentation design, not templates

snapdeck.app

1 points by bigmacfive 8 hours ago

Hey HN, I'm Minkyu Lee, co-founder of Snapdeck — https://www.snapdeck.app

I’m a founder from Seoul, South Korea We built Snapdeck because we got tired of spending hours in Keynote or Google Slides tweaking layouts that AI could fix in seconds. Now you can just describe your idea, and Snapdeck generates an entire presentation — designed, formatted, and ready to present.

How it works: - You type a short prompt like “Pitch deck for a coffee startup that focuses on sustainability” - Snapdeck builds a clean, minimal slide deck (titles, visuals, charts, and theme included) - You can edit anything — layouts, charts, images — right inside the app - Export as PDF (and soon PPTX)

Why we built it: As designers and founders, we realized slides haven’t evolved since PowerPoint 2003. Everything is still manual — but storytelling shouldn’t be. We wanted a tool that could think visually, not just fill text into templates.

Tech stack: We’re using a combination of open-source LLMs and custom layout agents. Each “agent” decides how to visualize your content — from layout to color to typography. The orchestration layer is modular, and we’re exploring open-sourcing parts of it soon.

What we learned so far: - People love fast iteration. Designers use it to brainstorm, not just to finish. - Most users don’t want to design — they just want to express ideas visually. - The hardest part isn’t the AI — it’s making users trust that AI can design tastefully.

We’d love your feedback. Try it here → https://www.snapdeck.app

If you want to dig into how we’re handling layout generation or multi-agent orchestration, I’m happy to share details in the comments.

Thanks for reading — and thanks to the HN community for the endless inspiration.

bigmacfive 8 hours ago

Thanks for checking out Snapdeck! I’m Minkyu, one of the founders (based in Seoul ).

One of the hardest things we faced was teaching LLMs visual composition — not just text. We had to build our own layout agent system that scores composition and color balance, and we’re experimenting with letting multiple agents “vote” on design quality.

If anyone here is curious about the architecture, dataset approach, or layout evaluation logic, I’d love to dive deeper — happy to share more technical details. Also, I’d really appreciate feedback from the HN design and ML folks on where this could go next.