StoryBridge

• 

  StoryBridge-Hero-Section-ThumpNail

Inspiration

281 million international migrants worldwide face an invisible challenge every bedtime: how do you tell your child a story when you speak different languages? Only 12% of third-generation immigrants retain their heritage language. StoryBridge was born from personal experience — as a Burmese engineer living in Paris, I wanted to build the bridge I wished I had growing up.

What it does

StoryBridge is an AI-powered bilingual storytelling companion that creates interactive bedtime stories bridging a parent's native language with their child's adopted language. Parents select their home language (20+ supported), choose cultural elements (festivals, traditions, landmarks), and the AI generates a 5-scene illustrated story in both languages with audio narration.

Children shape the narrative through interactive choices — typing or speaking their decisions using voice input. Each choice generates a new scene with fresh illustrations and bilingual narration, making every story unique.

How we built it

Multi-agent architecture using Google ADK:

• Story Architect (Gemini 2.5 Flash) — Creates bilingual story outlines with cultural context, runs through ADK Runner with session state so children's choices genuinely adapt the narrative
• Illustrator (Gemini 2.5 Flash Image) — Generates warm watercolor storybook illustrations with native interleaved text+image output via ADK
• Narrator (Gemini 2.5 Flash Preview TTS) — Creates bilingual audio narration via ADK Runner

Frontend: React 19 + TypeScript with a custom warm storybook design system (earth tones, no AI slop). Voice input via Web Speech API.

Deployment: Google Cloud Run with multi-stage Docker build (Node frontend + Python FastAPI backend).

Challenges we ran into

• Getting all three agents to run through ADK Runner with proper session state management
• Achieving native interleaved output (text + image in single generation) from the Illustrator agent
• Ensuring culturally authentic illustrations — the AI needed careful prompting to include specific cultural elements like thanaka face paint and Thingyan water festival scenes
• Bilingual TTS quality varies across language pairs

Accomplishments that we're proud of

• All 3 Gemini modalities working together seamlessly: text generation, image generation, and text-to-speech
• Interactive choices are real — children's input genuinely shapes the story through ADK session state, not pre-scripted branching
• The watercolor illustration style creates a cohesive, warm aesthetic that feels like a real storybook
• Voice input lets young children who can't type yet participate in story choices
• Live production deployment on Google Cloud Run

What we learned

• Google ADK's Runner + session state pattern is powerful for maintaining conversational context across multi-turn agent interactions
• Gemini's native image generation produces remarkably coherent illustrations when given cultural context
• Designing for multilingual families requires thinking beyond translation — cultural authenticity matters as much as language accuracy

What's next for StoryBridge

• Real-time voice conversation mode using Gemini Live API for fully spoken storytelling sessions
• Story library with Firestore persistence so families can revisit favorite stories
• Parent pronunciation guide highlighting key vocabulary in the heritage language
• Community story templates where families share cultural story frameworks
• Offline mode for bedtime without internet

Built With

• adk
• docker
• fastapi
• gemini
• google
• python
• typescript