StoryLives

Inspiration

My younger sister asked me to tell her a story, but she kept interrupting—"No, make the squirrel fly to the moon!" "Wait, give him a laser!" That's when it hit me: What if a story could actually LISTEN and adapt? What if kids could become co-creators instead of just listeners?

What It Does

StoryLives is an interactive storyteller where kids co-create magical adventures with Lumi, an AI narrator they can talk to naturally. Children can interrupt to change directions, and watch their choices generate unique illustrations in real-time, while creating a living storybook that adapts to their imagination.

How I Built It

• Frontend: React + Vite with Web Audio API for voice capture
• Backend: Python FastAPI with WebSocket connections for real-time communication
• AI/ML: Google's Gemini Live API for natural conversation + Imagen on Vertex AI for illustrations
• Cloud: Deployed on Google Cloud Run (serverless, scales to zero)
• Caching: Cloud Storage with intelligent prompt hashing to minimize API costs

Challenges I Faced

• Real-time interruptions: Making Lumi pause mid-sentence when a child speaks was tricky! WebSocket bidirectional streaming was the key.
• Image generation latency: Kids won't wait 5 seconds for pictures. Solved with parallel generation, and narrating while rendering next scene.
• Billing setup: The $10 prepayment hurdle was real, but now we're live!

What I Learned

• Gemini Live API is GAME-CHANGING for natural conversation
• Kids have WILD imaginations—our story graph needs to handle ANY direction
• Serverless doesn't mean costless—caching is your best friend

What's Next

• Add multiple story genres (space, dinosaurs, underwater)
• Let kids draw characters and have Lumi recognize them (vision input!)
• Deploy on more regions for lower latency worldwide

Built With

• cloud-storage
• docker
• fastapi
• gemini-live-api
• google-cloud-run
• imagen
• python
• react
• redis
• vertex-ai
• vite
• web-audio-api
• websockets