Music Mood Classifier & Playlist Curator
Led a 7-person team (ML, data, backend, frontend, business) to deliver an end-to-end system that classifies music mood from raw audio and curates contextual playlists (Deep Learning course project) Engineered the data pipeline: converted raw audio into normalized log-mel spectrograms in 3-second segments, mapped genre labels to 5 mood classes and handled class imbalance with inverse-frequency weighting Model reached 79.1% test accuracy (78.9% ± 0.9% in 5-fold cross-validation) vs. a 30% baseline; served through a REST API and an interactive web demo