Edit: Thanks to Spotify (more or less) deprecating their web api, it has come to my attention a key piece of functionality (saving the playlist direct to spotify) is no longer working. Bummer. I’m working on a fix but I wouldn’t hold your breath…
At the age of 40 I find it challenging to find new music. Sure, the Spotify recommendation engine isn’t terrible, but it does leave a bit to be desired especially if you want more control over recommendations. Recently, I’ve been using LLMs to augment my discovery process with good results. I was sharing this with my buddy Dave (who is far more into music than myself) and was surprised he hadn’t been using AI tools for music discovery which got me thinking about building an app layer for this purpose – which is now SonicAI.
I know the world doesn’t need another GPT wrapper but in the case of SonicAI, I felt it could be worthwhile. Integrating an LLM with Spotify, by itself, is pretty powerful. Add to this a more controllable (by the user) gpt4o powered recommendation system and we’re able to surface more dynamic new music suggestions than the “black box” algorithms used by the big music distribution companies.
Using SonicAI is easy. You simply put in a few bands you are into, and enter your age. Boom. You’ll get some recommendation that the system will directly dump into a Spotify playlist in seconds. If you want more control, I built that too through more advanced options such as being able to specify what you are really into, genres you love or hate, and more. Creating an account allows you to store your history, likes and dislikes in a database which I’ll in turn use to curate more new music you might like (this is like a baby version of RLHF I’m baking into the loop).
The coolest part is I was able to build this in well under 24 hours from inception to where it is right now. I recognize this is yet another app that will likely be relegated to the “toy app” bin, but I also know people area always looking for new music.
If you are looking for something specific to be added, or want a feature you don’t see here – as always reach out! jeff.brines@gmail.com
Cheers and happy listening.
Addendum Added May 5th – I’ve had a number of people reach out and ask “wait, isn’t this just what Spotify already does?”. Yes and no. Spotify has its own recommendation system which I’m sure every user is familiar with. However, their team took an entirely different approach than what I’ve built SonicAI on. As opposed to me bumbling about explaining the differences, here is a concise robot written comparison of the two systems…
Spotify Recommendation System
- Data Source: Collaborative filtering (what others like you listen to), audio analysis, behavior tracking.
- Pros:
- Scales effortlessly.
- Real-time updates from millions of users.
- Solid at surfacing mainstream or adjacent artists.
- Cons:
- Feels like a black box—low transparency.
- Repetitive; stuck in loops.
- Lacks deep context or nuance about why you like something.
LLM (e.g., GPT-4o)
- Data Source: Language model trained on broad musical knowledge, reviews, genres, context, vibes.
- Pros:
- Introspective—can factor in mood, era, obscure taste, lyrical themes, instrumentals, etc.
- Transparent—can explain why I’m recommending something.
- Flexible—can adapt instantly to new info (e.g., “I like CHVRCHES, but I hate The xx”).
- Cons:
- Doesn’t have your Spotify data (unless you give it to me).
- Not dynamically trained on your listening history or real-time behavior.
Bottom Line:
Spotify is passive, pattern-based, and scale-driven.
ChatGPT is active, context-aware, and insight-driven.