haru talk

• Inspiration: Addressing the cognitive friction of manual journal logging by replacing it with a calming bedside conversation.

• What It Does: Mapping the 3 distinct AI companion personas, multimodal low-latency audio stream, structured JSON- mode diary summarization (sentiment scores 0.0–10.0), retrospective calendar, and automatic KakaoTalk/Gmail cron notifications.

• How We Built It: Breaking down the Next.js 16 App Router, React 19, Tailwind CSS 4, Supabase Auth/PostgreSQL, OpenAI API suite (GPT-4o-mini, Whisper-1, TTS-1), Kakao Memo, and Resend APIs.

• Challenges & Engineering Fixes:

  • Resolving React 19 strict impure rendering in real-time preview simulations.

  • Optimizing serverless audio multi-hop latency by piping raw memory buffers and streaming binary chunked MP3 files.

  • Silent dashboard token-sync middleware to circumvent short KakaoTalk access token lifecycles during cron triggers.

  • Overcoming database write race conditions by trapping PostgreSQL unique index constraint conflicts ( 23505 ) and implementing self-healing fallback loops.

• Accomplishments & Learnings: Glassmorphism dark-mode aesthetics, granular RLS database policies, and serverless buffer management.

Built With

• and
• api
• cloud-host
• database
• language
• technology-shields/badges-categorizing-every-framework
• utilized