MyBudayaHub

MyBudayaHub addresses the challenge of documenting "intangible" heritage—practices like traditional cooking, weaving, and ritual dances.

Comment

The Story of MyBudayaHub 🏮

## Inspiration

The vibrant tapestry of our cultural heritage is fraying. While buildings and artifacts (tangible heritage) are physically preserved, the soul of culture—the recipes, the craftsmanship, the oral traditions (intangible heritage)—is disappearing as older generations pass away.

We wanted to build a bridge between the wisdom of the past and the digital language of the future. We asked ourselves: "How can we make cultural documentation as engaging as a comic book?" Our goal was to turn "old-fashioned" archive work into an interactive, AI-driven experience that excites younger generations about their roots.

## What it does

MyBudayaHub is an AI-powered documentation suite that transforms memories into visual guides.

  • Submit: A user provides a title, description, and photos of a heritage practice (e.g., traditional weaving, a ritual dance, or a family recipe).
  • Analyze: The system's Analyst Agent (Gemini 2.5 Flash) scans the uploads to extract visual metadata and cultural nuances.
  • Synthesize: The Scribe Agent (Gemini 3.1 Flash Lite) takes the user's story and the analyst's data to draft a 5-8 step educational guide.
  • Verify: The user reviews and polishes the steps to ensure absolute cultural accuracy (Human-in-the-Loop).
  • Visualize: The Artist Agent (Gemini 2.5 Flash Image) renders a professional-grade comic strip illustrating the entire process in a high-contrast, educational format.

## How we built it

  • Orchestration: We utilized a Decoupled Multi-Agent Architecture. The Google ADK was instrumental in managing the Analyst agent's lifecycle and memory.
  • Intelligence Stack: We leveraged a specialized trio of Gemini models:
    • Gemini 2.5 Flash: High-speed visual analysis of heritage artifacts.
    • Gemini 3.1 Flash Lite: Cost-effective, high-precision synthesis of narrative steps.
    • Gemini 2.5 Flash Image: High-fidelity artistic rendering with image output modality.
  • Frontend: A "Mission Control" Dashboard built with Next.js 16 and Tailwind CSS 4. We chose a dark-mode, neon-accented aesthetic to signify that cultural documentation is a high-tech mission for the future.
  • Backend: A modern FastAPI server that manages the state machine of heritage submissions, AI analysis, and asset serving.

## Challenges we ran into

  • Multimodal Data Flow: Passing the Analyst agent's visual findings to the Scribe agent required a robust "context-passing" logic to ensure the written steps matched the uploaded photos.
  • Comic Panel Consistency: Getting an image generation model to respect a "sequential panel" layout in a single image required extensive prompt engineering and constraint-based steering.
  • Real-time State Management: Building a UI that updates as the "Analyst" and "Scribe" finish their work, while waiting for user approval, required a carefully designed polling and status system.

## Accomplishments that we're proud of

  • True Human-in-the-Loop: We successfully implemented a workflow where AI does the heavy lifting, but the human maintains ultimate control over the cultural narrative before final rendering.
  • Premium Aesthetic: We moved beyond a simple CRUD app to create a dashboard that feels state-of-the-art and important.
  • Zero-Shot Artistic Direction: Achieving a consistent "halftone comic" style across different culture types (food vs. dance) without needing fine-tuned models.

## What we learned

  • The Value of Agent Specialization: We learned that breaking the task into "Analyst", "Scribe", and "Artist" roles produces significantly higher quality than asking one general model to handle the entire pipeline.
  • Prompt-Driven UX: We discovered that specific linguistic constraints (like "start with an action verb") are the most effective way to "program" image models.
  • Cultural Sensitivity: Building this project highlighted the importance of accuracy in documentation—leading us to prioritize the user-approval step as a core feature rather than an afterthought.

## What's next for MyBudayaHub

  • Multi-Page Chronicles: Moving from single-page comic strips to full-length "Heritage Books."
  • *Digital archive *: Build a repository for all uploaded cultural heritages
  • Vocal Archives: Integrating traditional music and dialect recording to capture the rhythmic and linguistic soul of the heritage.
  • Global Heritage Map: A public gallery where these AI-captured traditions can be pinned to a map and shared with researchers worldwide.

Built With

Share this project:

Updates