A first look at SoundSoar, my capstone project that uses machine learning to predict trending songs. In this kickoff I set up the app on AWS Lightsail, outline the trending feature, and share what worked, what did not, and what is next.
Introduction
Hi, I am Tori Grasso, a freelancer and master’s student building a project called SoundSoar. The goal is to predict which songs are likely to trend based on their audio features. It blends my love for tech and music and is aimed at creators, influencers, and small businesses that want to pick the right tracks for their content. In this first post I share how I set up the app, what the trending feature will do, the tools I used, and what I learned this month.
Feature development
SoundSoar analyzes song data and estimates the likelihood a track will trend. I am using the Spotify API to collect features and a machine learning model to make predictions. The plan is to turn those predictions into clear, simple insights that help users choose music with confidence. The app will include a friendly interface and dashboards that focus on practical actions.
Progress this month
I stood up the web application on AWS Lightsail and built the basic navigation and user flows. With the foundation in place, the trending feature is roughly 60% complete.
Model features selected
I finalized the initial set of audio features for the prediction model. This will evolve as I validate results and tune performance.
Next steps
The next phase is all about data. I will collect and process Spotify data to train and evaluate the model. After cleaning and labeling the dataset, I will experiment with a few algorithms, measure accuracy, and iterate until the predictions are stable enough for a first release.
Retrospective
What went right
- Deployed the app on AWS Lightsail and got the core navigation working.
- The trending feature has a clear scope and is more than halfway done.
- Focusing on predicting streams instead of sentiment made the project more realistic within my timeline.
What went wrong
- I lost time early by trying to include sentiment analysis. After talking with my advisor I narrowed scope to streams, which required a reset.
- AWS setup took extra effort for configuration and troubleshooting.
How I will improve
- Set tighter milestones and protect scope so the core features land first.
- Start data collection earlier to avoid delays in training and evaluation.
- Treat sentiment analysis as a future enhancement once the core model ships.
