Utanowa (うたのわ) — Song Circle

Inspiration

My mother has dementia. She can't recall song titles anymore — but she still hums the melodies from her youth, perfectly. I built Utanowa (うたのわ, "circle of songs") so that hum is enough.

What it does

1. Hum it — Hum ~10 seconds. Humming recognition identifies the song, with strong coverage of Japanese Showa-era pop and Western oldies: the music today's 70–90 year olds grew up with.
2. Hear it — A 30-second Apple preview plays instantly, in-app. No friction — critical when you have a 10-second window of recognition.
3. Talk about it — Claude AI generates a "Memory Card": a warm story about the era of the song, personalized to where and when the listener grew up, plus gentle open-ended questions. No quizzes, ever — we never test a person with dementia.
4. Keep it — Families log reactions (smiled, sang along, teared up) into a private album of rediscovered songs.
5. Personal Mode — A protected screen for the person themselves: one giant play button, zero billing UI, exits only via a 2-second long-press.
6. Care Circles (v1.1) — My mother moved away from her old friends. With an 8-character invite code (no account creation — a real barrier at 85), family and old friends join her circle, hear the songs she remembered, and share old photos and comments. Every comment AND photo is screened by Claude AI before anyone sees it — tuned specifically to protect the dignity of a person with dementia.

How we built it

React Native + Expo (EAS Build — no Mac needed), ACRCloud humming recognition, Apple iTunes previews, Supabase (Postgres + RLS + anonymous auth + Edge Functions), Claude AI (Haiku for memory cards and text moderation, Sonnet for photo moderation), and RevenueCat powering the family-side subscription (¥800/month, 7-day free trial). Solo developer pair-programming with Claude Code: 18 days from the first line of code to App Store submission, built fully in public.

Challenges we ran into

YouTube embeds are blocked by rights holders for most Showa-era songs — we rebuilt playback around Apple's 30-second previews. Our AI photo moderator once rejected an innocent manga page as "inappropriate" — we learned that moderation needs not just accuracy but a kind way of saying no. And Apple's review queue taught a solo shipper humility: our v1.0 is still awaiting final approval as the deadline passes.

Accomplishments that we're proud of

The first real test: my mother hummed "上を向いて歩こう" (Sukiyaki), the app found it, and she smiled before the preview finished. A dignity-first design shipped end-to-end: no quizzes, no billing on the patient's screen, AI that blocks demeaning comments before they are seen. And a verified RevenueCat purchase flow with a paywall redesigned to Apple's fairness standards.

What we learned

Build for one real user first — my mother — and the product decisions make themselves. Ship in public: every rejection and false-positive became our best content. And accessibility isn't a feature list; it's removing one barrier (account creation) so an 85-year-old friend can join a circle with an 8-character code.

What's next for Utanowa (うたのわ) — Song Circle

Rolling Care Circles out to my mother's real friends, Android support, and helping more families find "that one song" that opens thirty minutes of conversation.

Built With

• acrcloud
• anthropic
• claude
• expo.io
• postgresql
• react-native
• revenuecat
• supabase
• typescript