Inspiration — The gap between Gen Z journaling (77% try, 68% quit in 30 days) and the generic output of prompt-first AI music tools. The belief that everyone has music inside them, they just need a composer that listens to their life first. What it does — Diary entry + photo/video → AI composes 3 complete candidate songs with lyrics → user picks and publishes as producer. Covers both the artist side (diary-first composer) and listener side (follow by 감성/emotional vibe).
How we built it — TanStack Start + React 19 frontend, Supabase (Postgres/Storage/auth), Lovable AI Gateway with Gemini 2.5 Flash for the two-agent orchestration (Lyricist/Mood Capturer + Cultural/Musical Skill Matcher), text-embedding-3-small + pgvector for recommendations.
Challenges — Making lyrics personal not paraphrased; ensuring 3 genuinely different candidates; building recommendations from sparse early-user data; media/audio handling with signed URLs and fallbacks.
Accomplishments — End-to-end two-agent pipeline in <60s; mood-based recommendation feed; bilingual polished UI; 100% pick-and-publish rate in POC.
What we learned — Diary-first beats prompt-first; offering choice (3 candidates) transforms consumers into producers; vector search over artist embeddings is surprisingly expressive for mood matching.
What's next — iOS launch, Day One/Notion integrations, Producer tier monetization, branded mood-soundtracks for agencies, $200k ARR target.
Built With
- claude
- lovable
- sunoai
Log in or sign up for Devpost to join the conversation.