cultural-recommendation-app

The privacy-first concierge that knows what you’ll love before you do

Comment

CultuRank: AI-Powered Cultural Matchmaking

Inspiration

We wanted to bridge the gap between cultural tastes and personalized discovery—without compromising privacy. Why should platforms like Spotify or Netflix have all the fun? With Qloo’s Taste API and modern LLMs, we saw a chance to build a system that understands the why behind your preferences and connects them to unexpected recommendations—like suggesting a Lisbon boutique hotel because you love Wes Anderson films.

What It Does

CultuRank is a privacy-first recommendation engine that:

  • Analyzes your cultural tastes (e.g., "I love David Lynch films and Japanese jazz") via natural language.
  • Uses Qloo’s API to find semantic matches across domains (music → fashion, books → travel, etc.).
  • Generates LLM-powered explanations (e.g., "You’ll love this Berlin vinyl shop because your taste in X aligns with Y").
  • No personal data required—just pure cultural intuition.

How We Built It

  • Backend: FastAPI server calling Qloo’s Taste API (/entities/similar) and OpenAI’s API for explanations.
  • Frontend: Streamlit for rapid prototyping (or React for a polished demo).
  • Workflow:
    1. User inputs tastes → Qloo fetches cross-domain matches.
    2. GPT-4 generates a "cultural compatibility score" + reasoning.
    3. Results rendered as interactive cards (try it, save it, share it).

Challenges We Ran Into

  • Qloo API Learning Curve: Mapping abstract cultural preferences to actionable recommendations took iterative testing.
  • LLM Prompt Engineering: Getting concise, accurate explanations required tuning (e.g., avoiding hallucinations).
  • Latency: Balancing Qloo’s API calls + LLM responses for real-time feedback.

Accomplishments We’re Proud Of

  • Built a fully functional prototype in 48 hours with end-to-end integration.
  • Demonstrated cross-domain magic (e.g., connecting "vintage sneakers" to "hidden jazz bars in NYC").
  • Won Best Use of Qloo API at the hackathon (hypothetical, but aim high!).

What We Learned

  • Cultural data ≠ personal data: You can infer a lot from tastes alone.
  • LLMs are force multipliers for recommendation engines—they turn raw data into stories.
  • Simplicity wins: Our minimalist UI tested better than overloaded dashboards.

What’s Next for CultuRank

  • Domain-Specific Modes: "Travel Mode" (plan trips based on your music/film tastes).
  • Social Features: Compare taste profiles with friends (Qloo’s privacy-safe "taste twins").
  • API Expansion: Integrate booking links (e.g., Reserve a table at that recommended restaurant).

Demo: [Live Link] | Code: [GitHub Repo]

Built With

Share this project:

Updates