Polymath applies multiple neural networks to process music libraries at scale. It separates songs into stems (vocals, drums, bass, etc.) using Demucs, detects musical structure with sf_segmenter, identifies pitch and key with Crepe, and transcribes audio to MIDI using Basic Pitch. The processed output is quantized to consistent tempo and beat-grid, enabling music producers to rapidly search, sample, and recombine elements across their entire library into DAW-ready stems.