Sounds Like You — TUM Hackatune
A recommendation UI built on Cyanite's music similarity backend
At TUM Hackatune, our team built a UI on top of Cyanite’s music recommendation backend that lets a user pick songs from their liked tracks and get explainable recommendations back.
Users select seed tracks from their taste profile, and Cyanite’s similarity search finds matching tracks; an LLM turns Cyanite’s similarity tags into a plain-language explanation of why each song matches. A trait-ranking and AI-summary view breaks down which similarity tags (time signature, instruments, BPM, valence/arousal, genre, vocals) matched most, and which original song features were strongly preserved, partially preserved, or neglected in the recommendations. A mood-shift map visualizes how each recommended track drifts from the liked set along axes like calm/energetic and dark/uplifting. Finally, users can refine recommendations with a free-text prompt (e.g. “but I want something more up-beat”), with prior queries aggregated to keep guiding the recommendations toward the user’s preference.
