A web application that recommends songs based on user input songs and contextual signals such as mood or activity.
Why - Music discovery is broken for most people. You either rely on Spotify's black-box algorithm, which you have zero control over, or you manually hunt for new songs yourself. There's no tool that lets you say "find me songs that sound exactly like this" based on the actual science of what makes music sound the way it does.
Music recommendation systems are typically driven by collaborative filtering and large-scale behavioral data.
This project explores an alternative approach by prompting with a song and combining that with contextual inputs to generate recommendations.
Users provide:
- One or more songs they like
- A mood or activity
The system then generates recommendations based on:
- Audio feature similarity
- Context adjustments
This project demonstrates:
- Data ingestion and preparation
- Feature engineering
- Recommendation modeling
- Building a simple data product (web application)
- Potential enhancements:
- Integrate real-time Spotify API data
- Improve recommendation model with clustering
- Add collaborative filtering
- Improve UI/UX
- Add playlist generation
- Ecclesia Morain — Data Engineering, System Design
- Shital Rewanwar — Machine Learning & Recommendation Modeling