Music Recommendation System

A content-based music recommendation engine using Scikit-Learn, utilizing Cosine-Similarity and Z-Score Normalization to process 39,000+ high-dimensional audio feature vectors.

The system has a dynamic inference pipeline that solved cold-start issues, utilizing mean imputation to handle sparse user inputs, allowing the model to accept sparse user inputs (e.g., abstract mood scores from 0.0 - 1.0, floats) and dynamically reconstruct query vectors.

The system is currently designing a RESTful API backend (Flask) to serve model inference to a React web application and Chrome Extension, enabling real-time, low-latency query processing for end-users.