whatImeant.: Building an LLM-Driven Engine for Music Semantics Discovery

Music has been, since ancient times, a faithful companion to humanity, capable of enlightening, accompanying, and consoling. While its consumption was once constrained by the limited resources of past eras, today it is so readily accessible that it is often taken for granted. Although this pervasive accessibility has turned music into an omnipresent element of everyday life, it has simultaneously diminished the level of attention dedicated to it. Listeners frequently fall in love with melodies before lyrics, and many never venture into exploring a track's semantic depth. In a world defined by shrinking time and shorter attention spans—despite having the boundless archive of the web at one's fingertips—even simple tasks like searching for the meaning of a specific word, verse, or song feel cumbersome, discouraging curiosity and reducing discovery to a sporadic event. The purpose of whatImeant. is to make the exploration and contextual analysis of musical compositions—and the artists behind them—effortless and immediate, breaking down the barrier of user inertia. Conceived as a second-screen application, it enhances the streaming experience by pairing real-time playback data retrieval via official APIs with deep semantic analysis driven by Large Language Models (LLMs). From an architectural standpoint, the project is underpinned by an asynchronous backend engine specifically designed to manage the complexities of querying Large Language Models at scale. The system breaks down musical compositions into structured analytical layers—extracting overarching themes, historical and cultural influences, production references, and line-by-line rhetorical interpretations. To ensure robustness and low latency, the orchestration layer employs dynamic multi-model routing, prompt optimization techniques, and intelligent fallback strategies capable of gracefully navigating rate limits and service interruptions. Combined with a persistent caching mechanism that prevents redundant model invocations, the backend delivers rigorous, deep-tier semantic insights in real time, transforming raw lyric data into an engaging, interactive companion to the listening experience.