Audio Book Builder

App: https://ai-agent.galaxygate.app Try this input: An old harbor master counts the boats after a storm. Keep 10 pages, press Build the book, then press Listen on a finished page. Code: https://github.com/winbrignoli/ai-agent (folder demos/bookbuilder) WORKSPACE_ID=1526973045432684544 INSTANCE_ID=1551340429716578304 INSTANCE_NAME=hack-ai-agent APP_ID=1551342604035092480 APP_NAME=tts APP_DOMAIN=ai-agent.galaxygate.app APP_URL=https://ai-agent.galaxygate.app IMAGE=127.0.0.1:5000/hackathon:r1

Inspiration

The track ships two demos: one writes a book fast, one reads text aloud. We wanted a single app where the model does not just write the book but decides who should read it, and then you hear that voice on the pages as they land.

What it does

Type an idea. Kimi K2.6 on RunPod plans the chapters and, in the same reply, casts a narrator from the 20 voices of RunPod's Chatterbox Turbo endpoint, with one sentence on why that voice suits the book. The title page shows the choice the moment it is made. Pages stream in as they are written. Every finished page has a Listen button: the server sends the page to Chatterbox in the cast voice, saves the audio, and plays it. The finished book gets a shareable link where every page can be read aloud too.

How we built it

We started from the organizers' Bookbuilder starter (FastAPI, server-sent events, one request per page). We added a narrator field to the outline prompt with a lenient parser that falls back to a default voice, a narration module that reuses the Text to Speech starter's Chatterbox call (sentence chunking, wav joining, a per-page cache on the host volume), two narrate routes, Listen buttons on the live book and share pages, and player UI state.

Built With

• claude