VibeMatch

Language-free visual journaling powered by Gemini AI. Upload moments, get concept tags instead of words, discover similar vibes through AI-matched atmospheres. Visual storytelling.

Comment

💡 Inspiration

We live in a world drowning in words. Every moment needs a caption, every photo needs a description, every feeling needs to be articulated. But what if we could capture the essence of a moment without language?

VibeMatch was born from a simple question: Can AI understand the atmosphere of a moment better than words can express it?

Traditional journaling forces you to translate raw feelings into language—losing nuance in the process. We wanted to create a visual-first journaling experience where images speak for themselves, and similar "vibes" connect naturally through AI-understood concepts rather than hashtags or keywords.

🔨 What it does

VibeMatch is a language-free visual journal powered by Gemini. Here's the magic:

  1. 📸 Capture the moment - Upload a photo of your environment, mood, or scene
  2. 🤖 AI extracts the essence - Gemini analyzes the image and generates:
    • Objective visual descriptions (what's actually in the scene)
    • Abstract concept tags (木質/wooden, warmth, light, shadow, atmosphere)
    • Feature embeddings via Neuronpedia integration
  3. ✨ Discover similar vibes - Browse a feed of moments that share your atmosphere, matched by AI-understood concepts rather than manual tags

No captions needed. No forced descriptions. Just pure visual storytelling.

🧠 How we built it

Frontend & Backend:

  • Next.js 15 (App Router) + React 19 for the web interface
  • Tailwind CSS with custom glass-morphism components for a modern, atmospheric UI
  • Supabase (PostgreSQL + Storage) for data persistence and image hosting
  • Custom vibe-matching algorithm using concept overlap scoring

AI Pipeline (The Heart of VibeMatch):

  • Multimodal vision capabilities to "see" the photo
  • Generates objective scene descriptions
  • Extracts abstract atmospheric concepts
    • Neuronpedia API for feature activation analysis:
  • Identifies SAE (Sparse Autoencoder) features triggered by the image
  • Maps neural activations to human-interpretable concepts
  • Creates rich semantic embeddings for vibe matching
    • Python pipeline (vibeflow_mvp/) orchestrating the AI workflow: image → Gemini vision → description + concepts → Neuronpedia SAE → feature activations → Concept mapping → storable vibe profile

Built With

Share this project:

Updates