Introduction

As musicians, we often overanalyze when listening to our favorite songs. With nuanced audio features like time signature, key, and energy, we saw an opportunity to merge our passion for programming with musicianship by building a recommendation engine powered by AI.

What is Spinder?

Spinder takes a playlist and analyzes each track’s audio features to recommend new songs. The user can swipe right (like) or swipe left (dislike), just like Tinder. From there, Spinder curates a personalized playlist based on the user’s preferences.

Tech Stack

Frontend: Swift
Backend: Flask + SQLAlchemy
Database: 1.2M+ Spotify tracks with nuanced audio features
AI Integration: Gemini API for track recommendations

Workflow:

The frontend sends the playlist to the backend.
The backend queries both Spotify’s API and our track database.
Features are processed and sent to Gemini for recommendations.
Results are returned to the frontend for user swiping.
A curated playlist of liked songs is generated.

Challenges

Sending API requests from Swift
Adapting to long, intense hackathon hours
Collaborative development with Git
Spotify deprecating API features, which then forced us to pivot and integrate the ReccoBeats API

Achievements

Despite setbacks, we built a fully functional AI-powered music recommender in just one day. Models like this already shape many apps we use daily, and incorporating them into our very first project shows promise for future work.