L / 02Open lab

VoxDoc Vox

Voice-controlled presentations with semantic slide routing.

Next.jsEmbeddingsVoice

"Speaking and clicking are the wrong two things to do at once. So this stops asking you to."

IOverview

Why it exists.

Slide decks are designed to be presented but built like documents — you click forward, you click back, and you spend half the talk negotiating with a remote. Voice control sounds like the answer until you try the obvious version: 'next slide.' That doesn't help. You're not thinking in slide indices, you're thinking in topics.

VoxDoc routes navigation semantically. Each slide is precomputed with embeddings when the deck is generated. As you speak, the app matches what you're saying to the slide that fits — so saying 'let's look at the architecture' jumps to that slide regardless of where it sits in the order. You can also say 'next' if you really want to, and it works the way you'd expect.

Generation runs in the background. You provide context, walk away, and come back to a finished deck. Up to five saved presentations per user, sessions for re-opening across devices, and a downloadable PDF export when you need a handoff.

IIWhat it does

The surface.

01

Login sessions

Accounts persist across devices. Pick up a presentation where you left it.

02

Saved presentations (max 5 per user)

The cap is intentional — encourages curation over hoarding.

03

Background slide generation

Kick off a deck, walk away, come back to a finished one. Long-running jobs are first-class citizens.

04

Precomputed slide embeddings

Semantic routing works because the math is already done at build time.

05

Voice-driven navigation

Say what's next conceptually; the app finds the slide. Linear navigation still works for the simple cases.

06

Downloadable PDF export

For the handoff after the talk, or for archival.

IIIThe stack

Under the hood.

Frontend
Next.js
API
Node.js
Database
SQLite
AI
LLMsEmbeddings
Deploy
Render