Recently updated my open source app Back2Back to be multiplatform. Still all native code, but targeting macOS and iOS.

Use the default DJ persona's to explore different music, or create your own persona. Use GPT-5 in high thinking with web search tooling to create in depth style guides.

Each selection is made by an LLM using the persona, then another LLM matches the selections to Apple Music, and a final LLM validates that the Apple Music track fits the DJ persona.

This workflow provides the very best results, much better than other ML-based methods I have used in the past.

Still in beta, would love feedback.