B2Travel

B2Travel uses semantic AI to organize travel inspirations from your browser bookmarks and transforms them into an immersive 3D VR world you can explore through an interactive AI voice agent.

Comment

B2Travel 🌌✈️

B2Travel changes how you save, organize, and experience your travel inspirations. Instead of losing your favorite destinations in a messy camera roll or scattered browser bookmarks, B2Travel automatically organizes them using advanced semantic AI and projects them into a fully immersive, interactive 3D VR environment.

🚀 Features

  • Chrome Extension: Right-click and save any image or quotes straight to your Second Brain.
  • AI-Powered Semantic Clustering: Our backend utilizes OpenAI's CLIP Model to extract 512-dimensional semantic embeddings from your images. Snow mountains group with snow mountains, and sunny beaches automatically group with other beaches.
  • UMAP 3D Projection: We use UMAP (Uniform Manifold Approximation and Projection) to squash high-dimensional semantic spaces down into a visual 3D coordinate system.
  • Immersive WebVR Experience: Built with A-Frame, B2Travel renders your saved destinations in a beautiful, interactive Virtual Reality "Memory Universe." Look around, use gaze-based interactions, and travel through your memories straight from your phone or VR headset.
  • Conversational Voice AI Agent: Integrated with ElevenLabs Conversational AI, you can speak directly to a VR voice assistant. Say "Take me to a summer beach," and the AI dynamically calculates the semantic coordinates and visually guides you through the VR space to that exact vibe!
  • Multi-Modal Recommendations: Select photos you like in VR by looking at them, and the frontend instantly beams the images to your Voice Agent to provide personalized travel recommendations.

🏗️ Architecture (End-to-End Pipeline)

  1. Input (Chrome Extension): Click/Save an image or text while browsing.
  2. Backend (FastAPI): Receives the data and runs it through the CLIP-ViT model to generate embeddings.
  3. Storage (MongoDB Atlas): Stores the images, text, and vector embeddings.
  4. Dimension Reduction (UMAP): It maps all vectors to 3D space.
  5. Presentation (A-Frame / CodePen): A frontend WebXR interface fetches coordinates and images, rendering them into a beautiful, explorable VR galaxy.
  6. Multi-Modal AI Interaction (ElevenLabs): A Python agent continuously listens to the user's voice. When the user asks for a vibe, it emits a real-time event that the VR frontend polls and reacts to, actively guiding the user.

🛠️ Tech Stack

  • Backend: Python, FastAPI, PyTorch, HuggingFace Transformers (CLIP), UMAP-learn
  • Voice AI Agent: ElevenLabs Conversational AI API
  • Database: MongoDB (Atlas)
  • Frontend: A-Frame (WebVR/WebXR), HTML5, Vanilla JS, CSS
  • Tools: Chrome Extension APIs, ngrok (for tunneling localhost to mobile VR)

Built With

Share this project:

Updates