TrackShift

YouTube is where music gets discovered; Spotify is where it gets lived with. Moving a playlist between them is a slow copy-paste loop - search each song, confirm the right version, add it, move to the next. A 30-song playlist is a half-hour of fiddly work, and the titles on YouTube rarely match the catalog names cleanly.

TrackShift is a single form. Paste a link and the pipeline does the rest: extract the track list, clean up each raw title (stripping "Official Video", "Lyrics", quality tags, and everything else that keeps a title from matching), find each track on Spotify, and show you a preview of what matched before you commit. Connect your Spotify account and the playlist builds itself - in the same order, private by default, named after the source.

Next.js on Vercel serves the UI and the API routes. Track extraction runs with yt-dlp, so there's no YouTube API key and no quota to manage. The title-cleaning step uses an LLM on NVIDIA's free NIM endpoint - it turns messy YouTube titles into clean artist-song pairs that search well.

Spotify search and playlist creation use the standard OAuth flow: the app has its own registration, users authenticate with their own accounts, and nothing is stored. The pipeline stays inside a 60-second function budget, so it runs on a free Vercel tier with no database.

What took a half-hour of manual searching now takes one paste and under a minute. The preview shows exactly what matched before anything touches your account, so you never end up with a playlist full of wrong versions. Tracks that can't be found are flagged instead of guessed.