Inspiration

We all come from families where culture and tradition matter, and the way all of it gets passed along is by telling stories out loud. As we move away from home and make new friends, we lose touch with those stories, and we lose the language they're told in. A lot of us don't speak the same tongue as our grandparents, let alone our great-grandparents. The people holding the stories end up with nobody who can understand them. We wanted to fix that, because nothing beats a story told well: it takes you somewhere you've never been and lets you feel something you never lived.

What it does

Kinn is two pieces. Grandma has the Hearth, a framed screen on her kitchen shelf with no apps and no keyboard. She holds it for three seconds and tells a story in Spanish. I get it on my phone in English, spoken in a clone of her voice, as a film whose scenes change with her words, so none of the context gets lost. "Hear her original" plays her real recording.

It works both ways. In our demo I sent her a note about my weekend. She heard it in Spanish, in my own voice, over a film of my day, and answered: "Hi Angel, I hope I get to see you soon. I'll bring you some little cookies." Every story stays in the family archive so the next grandchild can hear it too.

How we built it

The Hearth: a web app running in our own Android kiosk on an Alexa Tablet 8.
The family app: SwiftUI on iPhone.
Backend: Supabase (Postgres, storage, realtime, edge functions).
Speech to text: ElevenLabs Scribe.
Translation: Two LLMs (GPT-6 Luna through a Codex relay) and Meta Muse (through api credits) translates with the warmth and idioms kept, writes a short title for every note, and suggests what to ask next.
Voices: ElevenLabs instant voice cloning, so each person is heard in their own voice in the other language.
Film: Luma ray-3.2 turns each beat of a story into a scene, with a soft, dreamlike film grain.
How we worked: three developers ran a fleet of AI coding agents (Claude Code and Codex) through GitHub issues and pull requests, with QA screenshots, end to end tests, and reviews between the two teams' agents.

Challenges we ran into

Grandma's real voice played back as a robot, because iPhones can't play the tablet's WebM recordings. We now convert every recording to m4a.
Voice cloning silently fell back to a stock voice: the API key was missing the permission to create voices.
The film lagged the narration. We timed every scene against the real audio, and playback now waits until the film is ready.
A layout loop froze the app on launch, and we only caught it on a real phone.
Doing all of it in 24 hours, on real hardware, in two languages.

Accomplishments that we're proud of

It works end to end on real devices, in both directions: a story recorded on the tablet, real translation, cloned voices, a generated film, and a reply that comes back. A grandmother never touches a keyboard, and her story arrives with her voice and her meaning intact.

What we learned

The AI only matters when the person stays in it: their voice, their words, their pace. The story is the product, and the technology's job is to get out of the way.

What's next for Kinn

Private family archives with real accounts, so every story is kept for the next generation.
Correcting a translation while keeping the original.
Remembering family names and sayings across stories.
A bilingual keepsake the family can print or share.
More languages.
One tap on any WhatsApp voice note.