Ins# TastePulse AI: Where Culture Meets AI Personalization

🌟 Inspiration

While scrolling through endless recommendation lists, I realized:

"What if AI could understand my tastes as well as my best friend does?"

The spark came from two observations:

  1. 90% of recommendations feel generic (cough Netflix algorithms).
  2. Cultural tastes follow hidden patterns—your love for Bengali rock might predict your interest in Kolkata’s indie cafes.

🛠 How I Built It

Tech Stack

  • AI Brain: Gemini + Qloo API (for cultural entity mapping)
  • Math Behind the Magic:
    Taste matching uses cosine similarity:
    $$ \text{sim}(\mathbf{u},\mathbf{v}) = \frac{\mathbf{u} \cdot \mathbf{v}}{|\mathbf{u}| |\mathbf{v}|} $$ where $\mathbf{u}$ and $\mathbf{v}$ are user taste vectors.
  • Frontend: Vue.js + Netlify (lightning-fast UI)

Key Features

  1. Multilingual Parsing
    • Understands "আমি রবীন্দ্রসংগীত পছন্দ করি" → recommends Tagore-inspired travel.
  2. Cross-Category Links
    • "I love cyberpunk books" → suggests synthwave playlists and Tokyo neon spots.

🧠 Lessons Learned

  1. Latent Preferences Matter
    • Users who like Haruki Murakami often enjoy jazz bars (even if they don’t explicitly say so).
  2. Privacy ≠ Personalization Tradeoff
    • Achieved 85% accuracy without storing personal data (only taste fingerprints).

⚡ Challenges

  • API Limitations: Qloo’s "comedy" tag included dark humor (required manual filtering).
  • Bengali NLP: Had to build a custom transliteration layer for রবীন্দ্রসংগীত → Rabindrasangeet.

🚀 What’s Next?

  • Real-Time Taste Adaptation using RNNs:
    $$ h_t = \sigma(W_{xh}x_t + W_{hh}h_{t-1} + b_h) $$ to predict evolving preferences.
  • Localized Partnerships (e.g., Dhaka’s best হোটেলের খাবার recommendations).

"Culture is chaos—we’re here to make it click." piration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for TastePulse

Built With

Share this project:

Updates