Global Music Recommendation Using Deep Contrastive Learning and Hybridized Filtering

Developed a hybridized music-recommendation system combining deep contrastive learning with content-based and collaborative filtering, achieving a 24.2% performance gain (AURC 296.52 vs. 238.81 / 142.90) over CF/CBF baselines.

Trained a Triplet-Loss deep network on 170,000 triplets from 13,130 playlists and engineered a language-agnostic model able to recommend Hindi songs to English-language users via learned embeddings. Integrated the Spotify Web API and tuned the model with Bayesian Optimization, improving Recall@K by 27.6%.