AI Product & Music Information Retrieval
Resonant
From natural-language music discovery to private DJ mixes
Overview
Resonant is the music application I am developing. It connects a natural-language request to an explainable playlist, then to a planned DJ mix with private progressive playback. The current AWS deployment is a bounded private trial, with accounts and persistent playlists and jobs.
Retrieval that can be inspected
PlaylistIQ translates a request into validated musical intent, retrieves tracks using semantic embeddings and musical features, and applies deterministic ranking. Each track has an explanation, and fixed-query replay checks help preserve behavior as the system changes.
Acoustic analysis runs locally. The cloud uses exported catalog evidence and a focused text-inference worker for planning and retrieval.
From playlist to playback
The playlist-to-mix flow creates versioned transition plans and renders progressive AAC/HLS audio with completed M4A output. Owner-scoped durable jobs, idempotent admission, cancellation and fenced completion support reliable processing. Private media delivery isolates each account's results.
A smaller rendering service
I reduced the compressed renderer container from 310 MB to 73 MB with a focused FFmpeg build and lazy optional dependencies. The renderer remains separate from model inference, with source and artifact verification and bounded transition behavior.
Current state
The private trial supports discovery, saved playlists and progressive mixes. General access, personal cloud uploads and cloud acoustic analysis are still in development. A walkthrough is available on request; the live app requires an approved account.