Sonar

Ping an artist or song to discover what's nearby

Comment

Inspiration

I struggle with finding new music, so I wanted to make an app that could assist people in doing so. I also decided to make it able to accept either a song or an artist to have a bigger range of possible uses.

What it does

  • Search by artist or song: Sonar finds it on Spotify and shows a profile with the photo or album cover, genre tags, a short bio, and Last.fm listener and play counts.
  • Get recommendations: five similar artists or songs, each with a match bar showing how closely it relates to your search.
  • Go down the rabbit hole: click Explore on any recommendation to make it your new search. A clickable trail (Frank Ocean → Steve Lacy → The Internet) tracks your path, so you can jump back to any step.
  • Listen right away: embedded Spotify players let you preview an artist's top tracks or a single song without leaving the app.

How I built it

Sonar is written in Python with Streamlit for the interface, and it combines two APIs:

  • Spotify Web API (through Spotipy) to find the artist or song you typed, plus images, album info, links, and embedded players.
  • Last.fm API for the recommendations, match scores, listener and play counts, tags, and bios. Streamlit reruns the whole script on every click, so I used st.session_state to remember the current search and the exploration trail between reruns. I also cached every API call with st.cache_data, so repeat searches load almost instantly and use fewer API calls. The app is deployed on Streamlit Community Cloud, with the API keys stored as secrets.

Challenges I ran into

  • Two APIs name songs differently. Last.fm would suggest "Ghost Town (feat. PARTYNEXTDOOR)", but on Spotify it's just "Ghost Town", so the lookup found nothing and the card had no image. I added a fallback that retries without the "(feat. ...)" part, but only when that actually changes the title, so other searches don't make extra calls.
  • Results disappearing. Because Streamlit reruns everything on each click, adding Explore buttons at first made the results vanish. Learning session_state fixed it.
  • Scroll position. After clicking Explore at the bottom of the page, the new artist's profile loaded off-screen above. I added a small script that scrolls back to the top after each Explore.
  • Deployment. My first deploy crashed because the API keys are in a local .env file that isn't on GitHub. I had to add them as secrets in Streamlit Cloud. I also found my requirements.txt had been saved in the wrong text encoding by PowerShell and had to re-save it as UTF-8.

Accomplishments that I'm proud of

  • Building a complete, deployed app that anyone can try at sonar-music.streamlit.app
  • The Explore rabbit hole with its trail, which turns one search into a chain of discoveries
  • Combining two different APIs into one smooth experience, with fallbacks for missing images, empty bios, and mismatched song names

What I learned

  • How to work with REST APIs, API keys, and error responses
  • How Streamlit apps work: reruns, session_state, caching, layouts, and callbacks
  • How to deploy a Python app and manage secrets safely

What's next for Sonar

  • "Why you'll like this": AI-written explanations of what connects each recommendation to your search, using the tags and bios Sonar already fetches
  • Vibe search: describe what you want in plain English ("chill songs for a late-night drive") and get matching music
  • Clickable genre tags to explore top artists for a genre like "neo-soul"
  • Journey summary: turn your exploration trail into a playlist idea

Built With

Share this project:

Updates

Submission history