MUSES

Mobile App Screen - Delete Account function in Profile screen
Mobile App Screen - Details of Recommended songs to learn
Mobile App Screen - Study Plan to learn Recommended songs

Inspiration

Our team wanted to address the Education Track Challenge presented by Zoom— in particular addressing music education. As avid lovers and musicians, all 3 of us in the team wanted to join in the process of gamifying education by using AI algorithms to aid learning an instrument or vocal singing.

What it does

Our app will first display a tailored feed of music recommendations. They may choose to like, save, or scroll past each piece of music. The app will process data from likes, saves, and retention in a cosine similarity algorithm to determine musical repertoire recommendations for the user to practice and play.

How we built it

Our team used Flutter for frontend development and SQL relational database for backend development. With Flutter, we ran both an IOS and Android simulator to see the UI / UX design element changes in real time. The SQL database was used to store all the songs and musical pieces in our mobile app, while OpenAI API was used to generate vector embeddings for our recommendation engine’s cosine similarity search. Hence, our tech stack included Dart in addition to algorithms in Python for data ingestion and C++ for app development.

Challenges we ran into

We didn’t know how to create a recommendation system to match users with songs they might like, for which we want into challenges while generating and storing vector embeddings. In addition to navigating vector databases, our team struggled with merging the git commits and push-pull requests as working in a team in a hackathon was a new experience for all of us.

Accomplishments that we're proud of

The queue shuffler for songs on the home page of the mobile app is very enjoyable, as our AI LLM powered recommendation algorithm presents these songs to our users based on their favorite genres, previous liked and saved songs. Users have the ability to both save and like the songs recommended by our queue shuffler. The app also collects data from these interactions to fuel the recommendation algorithm.

What we learned

All of us in the team learned how to code in Flutter and FlutterFlow for frontend mobile app development, which has been a very enjoyable and learning experience for us. From possessing the added flexibility to drag and drop design elements to our OS simulator to learning how to create complex screens with embedded media links, learning to code in Flutter has been the highlight of Treehacks for us.

Additionally, we were able to utilize the new IDE developed by Codeium, Windsurf, whose Claude Sonnet-powered copilot has helped us brainstorm and plan complex algorithms.

What's next for MUSES

We have already started creating customer surveys for fellow hackers to gather user testimonials and interest. Based on the vast majority of our users listening to music through Spotify and YouTube Music, we plan to integrate Spotify song links into our mobile app’s homepage in addition to the YouTube Music integration. In the future after the hackathon ends, we plan to iterate MUSES and pivot if necessary to reach a bigger audience and spread the joy of learning music.