Can AI replace BulkPublish?

Build me a personal bulk social scheduler to replace BulkPublish. Requirements:

- Python 3.12 + FastAPI + SQLite, Jinja templates, no JavaScript framework.
  One process on an always-on box, APScheduler ticking once a minute.
- Two networks only: Mastodon (access token) and Bluesky (app password).
  One small module per network behind a single publish(post, account)
  function so a third is addable later. Scaffold nothing I do not use.
- The main screen is a CSV importer. Columns: caption, image, send_at,
  accounts. Show me every parsed row with a per-network character count and
  a red flag on anything too long or missing a file, and import only the
  rows that pass.
- Rows with no send_at fall into posting times I configure per account
  (weekdays 09:00 and 17:00 in my timezone) and take the next free one.
- Each tick picks up due rows one at a time and flips the row to "sending"
  in its own transaction before any network call, so a crash or a double
  tick can never post the same row twice. Then "sent" with the live post
  URL, or "failed" with the error text. Three retries, backing off, and
  never a retry once a send is confirmed.
- Images copied into media/ beside the database, checked for type and size
  at import, removed 90 days after the row sends.
- A history page: last 200 rows, status, error, link to the live post, and
  a retry button.
- Out of scope, and say so plainly in the README: X, Instagram, LinkedIn
  and TikTok (paid, approval-gated or review-gated APIs), analytics,
  approvals, and more than one user.
- README covers getting a Mastodon token and a Bluesky app password.
  Secrets in .env, localhost only, no accounts, no telemetry.