Sporkify

The Market Gap

Many websites and Spotify playlists claim to provide the perfect running/rowing/cycling music experience by matching songs to specific tempo or beats per minute (BPM). While these solutions offer some convenience, they often fall short for serious or personalized use. Most playlists feature generic, outdated pop tracks—often from the 1990s—that fail to reflect individual music preferences. For users who value their personal music library and want an experience tailored to their taste, these generic solutions are far from ideal. The market currently lacks a solution that blends accurate BPM matching with personalized music selection, leaving a significant gap for innovation.

Problem with existing BPM data

In essence, the idea behind the software would be to track/ask the user for a BPM target range based on their running, and find music that they listen to from their Spotify or Apple music account. However, the main problem that arises is in the current BPM databases.

Reputable BPM databases like Tunebat charge exorbitant fees (around $250/month) for API access to BPM data. This pricing model is prohibitive for independent developers like myself.

The alternative approach

To overcome these barriers, I explored an unconventional solution: using LLMs like Gemini 2.0 Flash and 2.5 Flash-Lite to search BPM data for songs. By providing the model with track names and asking it to ground its answers on a Google search, I was able to achieve surprisingly accurate results—about 96% accuracy in testing. This experiment demonstrated that public, AI-driven approaches can rival expensive proprietary databases for certain use cases. It also validated the potential for AI tools to democratize access to music metadata, opening new possibilities for personalized music apps without relying on costly or restrictive APIs.

Prototype Walkthrough

Spotify login

BPM or SPM preference

Showed animations of different running cadences as most users probably won't be able to count BPM.

Song sourcing preference

Playlist duration

Learnings and Drawbacks

While this experiment showed promise, it also revealed significant challenges. Running LLMs continuously is expensive: I spent roughly $180 in cloud credits in a single day (thankfully I didn't need to pay as I had free trial resources). Additionally, API restrictions from platforms like Spotify remain a major hurdle. Spotify impose limitations on API access, granting non-developer use only to apps with over 250,000 monthly active users.

Potential Next Steps

Looking ahead, there are several promising directions for development. One is integrating cadence data from fitness devices like the Apple Watch, which tracks steps per minute (SPM), to dynamically match music tempo with running pace.