Can AI replace Frame.io?

Build me a video review and approval app to replace Frame.io for a one-person edit shop. Requirements:

- Node + Express + better-sqlite3 running on my own VPS behind an existing reverse
  proxy. Server-rendered pages, one vanilla JS player, no frontend framework.
- I drop a master into ./incoming/<project>/<asset>.mov; chokidar picks it up and
  ffmpeg writes a 1080p H.264 proxy and a poster frame into
  ./media/<project>/<asset>/v<N>/. Frame rate comes from ffprobe.
- Versions stack: the same asset name dropped again becomes v2, v3, and so on. The
  review page gets a version dropdown that swaps the source at the current playhead.
- Each asset has one unguessable /r/<random-id> link. A reviewer types their name
  once into localStorage, then comments; every comment stores the exact frame and
  renders as HH:MM:SS:FF at the asset's real frame rate.
- Comments list beside the player and tick along the scrubber. Clicking one seeks to
  that frame. Each has a reply thread, a resolved checkbox, and an unresolved filter.
- An export button per version writes notes.csv (timecode, author, comment, resolved)
  and a DaVinci Resolve marker .txt I can import onto the timeline.
- SQLite holds comments and versions, media stays in plain folders, so nothing is
  trapped in the app. No accounts and no telemetry: admin pages sit behind basic auth
  with the password in .env, share links are the only other way in.
- Out of scope: Camera to Cloud, NLE panels, watermarking, and a mobile app. Do not
  build accounts, teams, or roles. The unlisted link is the entire permission model.
- README: installing ffmpeg, raising the proxy upload limit, and an honest warning
  that you now own the storage bill, the egress bill, and the 2am playback outage.