AUVIO
Sing a melody, get a produced song. Flutter app with a custom C++ FFI audio engine and three managed-GPU AI models on the backend.
An iOS and Android prototype that turns voice or humming into a fully produced track in under two minutes — vocals, drums, bass and instruments, all AI-generated, mastered, exportable. Built for hobby musicians who want song-idea-to-finished-track without ever opening a DAW.
01 · Approach
A meditation app that happens to make music.
The visual decision was the central one: iridescent + aurora instead of classic DAW chrome. Mood-board references are Calm and Endel, not Logic Pro. Soft gradient pairs (pink ↔ orange, cyan ↔ violet, yellow ↔ pink) become recurring magical objects.
Layout follows a single primary action per screen. The Liquid-Glass bottom-nav is custom: backdrop-filter blur, animated iridescent active pill, safe-area-aware.
Palette
Five colours, infinite gradients.
Stack
A two-stage audio pipeline.
- 01Flutter 3.x
- 02Riverpod 2
- 03C++ FFI Engine
- 04Replicate AI
- 05Supabase Postgres
- 06Celery
Margin-safe by construction
Every GPU call hits a budget guard.
A live RevenueCat check plus a monthly cost-budget guard means a free-user request cannot accidentally burn budget. Three Replicate models, wired end-to-end with status polling and FCM push.
Visit Tier Architecture →
Prototype, locked.
