Help in a Bottle

Generic AI gives cold advice. Help in a Bottle uses Gemini to match your struggles with wisdom from people who've been there. AI doesn't write the answers; we just find the human match.

Comment

Inspiration

Isolated people often fall through the cracks. Generic AI gives cold data; I wanted to use AI to facilitate genuine human connection and empathy.

What it does

It’s a digital "message in a bottle" platform. Users cast "bottles" of lived experience or seek help for current struggles. The app uses AI vector matching to pair users' emotional needs with real human wisdom, wrapped in a calming UI designed to reduce stress.

How I built it

  • Frontend: Next.js SPA, React 19, Tailwind CSS 4 Deployed on AWS Amplify.
  • Backend: Firebase Firestore.
  • AI Engine: Gemini (gemini-embedding-001 for semantic matching, gemini-3.1-flash-lite-preview for therapeutic synthesis).

Challenges I ran into

Learning to use Firebase for user data storage for the first time was a major hurdle. Specifically, figuring out how to securely store and query high-dimensional AI vector embeddings within Firestore was a steep learning curve. On the frontend, handling asynchronous data fetching while keeping the cinematic SPA UI perfectly smooth required complex React state management.

Accomplishments that I'm proud of

Shipping a fully functional, end-to-end MVP. I successfully connected a complex AI backend (vector math and semantic search) with a deeply calming, intuitive frontend.

What I learned

I learned how to map human emotion mathematically using Gemini embeddings to build a superior search experience. I also leveled up my frontend skills, mastering the Next.js App Router and advanced CSS blur techniques.

What's next for Help in a Bottle

Implementing voice-to-text and multi-language support for better accessibility, and partnering with community organizations to seed the database with verified "bottles" from support group leaders and professionals.

Share this project:

Updates