Spotify MoodMixer: Building an AI-Powered Playlist Curator

Background: The Struggle with Background Music

Like a lot of people, I love having background music while doing everyday things, whether I am working on assignments, meal prepping or just enjoying the sunset with coffee.

But the problem?

1. Finding the right playlist takes a lot of time.
2. Spotify’s playlists are great, but they’re not truly personalized.
3. Sometimes, I don’t even know what to search for.

I was working on a case study on Spotify’s market segmentation when I went deeper into this personalization gap and that’s when I decided to build something myself.

The Idea: What If AI Could Do It For Me?

I wanted to build something that curates a playlist based on what I feel, instead of me picking from pre-made playlists.

This became Spotify MoodMixer, a tool that:

• Understands natural language input (e.g., “I’m in a coffee shop in Paris” or “I am cooking pasta at night”).
• Uses AI to curate a new playlist from scratch (instead of picking an existing one).
• Saves it directly to Spotify so it’s instantly playable.

Handling Complex Inputs Like a Human Would

Unlike typical playlist recommendations that require specific genres or artists, Spotify MoodMixer understands complex, freeform mood descriptions and turns them into playlists.

For example, instead of searching for “chill music” or “upbeat pop”, users can type:

• "I had a long day and I missed my bus." → AI interprets tired, frustrated, and needing comfort → A mix of soothing acoustic, mellow jazz, and soft indie.
• "Futuristic but soft." → AI picks ethereal electronic, ambient synth, and dreamy soundscapes.

This level of emotional and contextual understanding makes MoodMixer feel like a personal DJ, curating the exact vibe you need in the moment.

How is This Different from Using Spotify's APIs?

Spotify already has APIs for searching songs and generating playlists, but here’s what it doesn’t do:

• It doesn’t interpret mood inputs in natural language. You still have to manually search by genre or keywords.
• It doesn’t generate a brand-new playlist every time. Spotify’s AI is historical (based on your past listens), not contextual (based on how you feel right now).
• It doesn’t let you discover new music intuitively. If you ask for "sunset music," you won’t get an AI-curated mix; you’ll have to scroll through existing playlists.

Spotify MoodMixer interprets what you type, picks songs dynamically, and builds a playlist for you.

Building the First Version (With Replit!)

I didn’t have a clear plan, just an idea. So, I fired up Replit to see what I could make.

Step 1: Prototyping with AI

I started by chatting with ChatGPT to brainstorm how I should structure the tool. The first thing I needed was a good prompt that replit could understand. I made the prompt such that it captures enough information for replit to build a foundational structure and then iterated over it to get to the results (Check Appendix for detailed prompt). Replit generated an initial structure, and from there, I started tweaking.

It was surprisingly easy to set up -within a few minutes, I had a basic frontend ready.

Step 2: API Setup (And Running into Limits)

• At first, I used OpenAI’s API, but I ran into quota limits fast.
• I had to switch to Cohere.ai because it let me use the API many more times.
• Getting the API keys and authentication set up took a few tries, but Replit made it easy to debug things in real time.

The best part was that I could tweak my AI setup like I would give feedback to a dev team, but instantly.

• If I didn’t like a certain AI response, I just adjusted the prompt.
• If I wanted a different playlist style, I regenerated in one click.
• If something broke, I didn’t have to wait. I fixed it myself in Replit.

This speed of iteration was a huge learning -building AI-powered tools today is way faster than before.

The Final Build: How It Works

TL;DR Summary:

1. User enters a mood or scenario→ AI interprets emotions.
2. AI selects a playlist theme→ Mood is translated into a search query.
3. Spotify API searches for matching songs→ AI filters and refines the song list.
4. A new playlist is created→ Tracks are added programmatically.
5. The user gets a personalized playlist.

This whole process takes less than a minute - no searching, no manual curation, just instant personalized music.

(If you want a detailed technical breakdown, check the Appendix section at the end.)

The Results & What I Learned

?? A demo of what I build -

? I built a working prototype in just a few hours.

? I needed AI knowledge but not extensively.?

? I could iterate in real-time, skipping traditional dev cycles.

What didn’t work as well?

• The AI song selection isn’t perfect yet. Cohere’s model(free version) isn’t as good as prominent AI solutions, but it’s fine for free API access.
• Spotify’s API has limits - setting up OAuth authentication took extra work.

Conclusion: The Bigger Picture

As I worked on Spotify MoodMixer, I realized that Spotify itself is already moving toward AI-powered playlist curation. Spotify recently introduced AI Playlist (Beta), a feature where users can describe a playlist in text, and AI will generate one for them. Unlike my tool, Spotify’s AI Playlist works within their app and uses more advanced AI models for song selection.Third-party tools like Playlistable and PlaylistAI also allow users to generate AI-driven playlists.

If I had researched these solutions before I started, I might have approached MoodMixer differently. But building it from scratch was still an eye-opening lesson in the power of AI in prototyping and how quickly ideas can go from concept to execution. I am unsure about the future of vibe coding but AI is definitely good enough at building initial prototypes.

Even though there are other solutions available, I still believe there is room to refine personalized music curation, and this project helped me understand that AI can be a great tool for quickly testing ideas(and even failing fast).

What’s a unique mood or scenario you’d want an AI to create a playlist for? Post it in the comments and I will try it out!

APPENDIX