Vinder

Scan any vinyl record, get instant info, recommendations and lore.

Comment

💿 Inspiration

We took one look at HackRPI's theme and realized we had to do something related to vinyl records. Nothing beats going to a record store and entering a time machine where old, inconvenient technology is cherished. But we identified a problem: when people go to record stores, there are typically tens of thousands of records to browse through. We wanted to optimize the process of picking out a record eliminating the need to manually look up prices to ensure they're fair, search Spotify for the most popular songs off that album, and discover similar artists you might enjoy.

What it does

Vinder is a vinyl record scanner that uses computer vision and AI to revolutionize the record shopping experience. Simply scan any album cover with your phone's camera, and Vinder instantly provides:

  • Smart Price Checking: Real-time market prices from Discogs to ensure you're getting a fair deal AI-Powered Recommendations: Similar artists and albums based on Last.fm's recommendation engine, with match percentages
  • Top Tracks: The most popular songs from the album so you know what you're getting
  • Spotify Integration: One-tap access to play any track, artist, or album directly in Spotify
  • Artist Insights: Biographies, listener stats, and genre information
  • Community Data: Ratings, reviews, and how many collectors want or own each album

All of this happens in seconds using built in computer vision just point, scan, and discover!

How we built it

We built this using Android Studio, and a lot of data scrapers,

Challenges we ran into

  • The Data Challenge: Getting quality training data for our computer vision model proved to be our biggest hurdle. Existing datasets are 200+ gigabytes—far too large to download and process within a hackathon timeframe. Our solution? We wrote custom Python web scrapers to pull album cover images from Spotify's top 750 artists, creating a targeted dataset of the most popular albums that users are likely to encounter.
  • Mobile ML Integration: Translating Python-based machine learning code and pip modules to run efficiently on Android was complex. We had to learn the intricacies of TensorFlow Lite, ML Kit, and on-device inference to make real-time scanning possible without draining battery or requiring constant internet connectivity. API Rate Limiting: Coordinating multiple API calls (Discogs, Last.fm) while staying within rate limits. ## Accomplishments that we're proud of ##
  • Multi-API Integration: Successfully orchestrated three different APIs (Discogs, Last.fm, Spotify) to work together, providing a complete music discovery ecosystem Real Computer Vision: Built a working ML-powered scanner that actually recognizes albums in real-time, not just a proof of concept Complete Feature Set: In 24 hours, we built features that apps charge subscription fees for and made it free Team Collaboration: Each team member contributed their unique skills, from API integration to UI design to ML model training, demonstrating excellent teamwork under pressure and a time limit. ## What we learned
  • Mobile ML is Hard: Deploying machine learning models on resource-constrained devices requires completely different thinking than server-side ML
  • API Design Matters: Well-documented APIs (like Discogs and Last.fm) saved us hours compared to poorly documented alternatives User Testing is Crucial: Testing with real vinyl records in different lighting conditions revealed edge cases we never anticipated ## What's next for Vinder

We plan on releasing this to the google play store once we are able to train the AI model better and work out some kinks. There is definitely an opportunity in the market to improve upon apps similar to this that already exists, as these are typically paid services for people keeping track of records in a collection, not people looking to learn more about their records.

Built With

Share this project:

Updates