Audio normalization pipeline

Design the audio post-processing pipeline for a music/podcast platform that, on every upload, must (a) measure and normalize loudness so tracks in a playlist don't jump in volume, (b) generate the visual waveform/peaks data the player renders for scrubbing, and (c) detect silence/clipping issues, for 100K uploads/day, with creators uploading anything from a quiet podcast to a heavily-mastered single. Normalization must be reversible/parameterized (don't destroy the original) and the waveform must render instantly in the client.