Suzy AI

Real-time AI companion with Live2D avatar, voice conversation, and emotion-aware animation.

Why This Project

I built Suzy as a safe, non-judgmental companion for people who feel lonely, socially awkward, or misunderstood.
Goal: let users talk naturally in any language and feel emotionally supported.

Standout Features

• Real-time voice chat using ElevenLabs conversational API.
• Live2D character rendering with lip-sync, gaze tracking, and idle animations.
• Emotion + gesture-driven visual feedback during conversation.
• Companion-style UX: onboarding profile, daily mood check-in, continue-chat flow.
• Responsive UI optimized for desktop and mobile.

Tech Stack

• React + TypeScript + Vite
• Tailwind CSS + shadcn/ui
• PixiJS + pixi-live2d-display
• ElevenLabs (voice)
• Supabase Edge Functions (session token backend)
• Vercel (deployment)

Architecture (Essential)

src/
├── pages/
│   └── Index.tsx                      # main conversation and UI orchestration
└── components/
    ├── CharacterDisplay.tsx           # model switching + loading/fallback behavior
    └── Live2D/                        # avatar rendering and animation pipeline

supabase/
└── functions/
    └── elevenlabs-session/
        └── index.ts                   # signed session URL generation

Quick Run

npm install
npm run dev

Create .env.local:

VITE_SUPABASE_URL=your_supabase_project_url
VITE_SUPABASE_ANON_KEY=your_supabase_anon_key