musicRUT

For my thesis, I created musicRUT, a proposed music-recommendation feature for Spotify designed to help listeners break out of their algorithmic ruts and avoid the hyper-personalization that keeps them hearing the same songs on repeat. In my user research, 90% of participants said Spotify repeatedly recommends the same artists, and 100% wanted at least some additional control. A few themes rang true: users want discovery to feel more human, more emotional, and more connected to the creative worlds they love.

I built the musicRUT MVP to give listeners simple, intuitive ways to influence their recommendations without adding cognitive load. Users pick a song they love, refine it with a few quick choices, and can optionally draw inspiration from film or photography. A hybrid similarity model powered by LLMs, VLMs, and external APIs then generates a personalized playlist, a short summary explaining why the playlist fits their selections, and AI-generated cover art. Even in MVP form, it surfaced artists Spotify had never shown me, proving how a bit of user agency can interrupt repetitive loops.

Building this project showed me how small moments of user control can make discovery exciting again. It also points toward a direction platforms like Spotify could explore: transparent, collaborative personalization grounded in UX rather than pure automation. With stronger models, richer data, and proper engineering resources, the system could grow far beyond its current form. Below is an early MVP prototype I designed in Figma with AI-assisted prototyping tools.