Faut-il créer votre propre Feedspace ?

Build me a testimonial system for my own site where the wall keeps itself up to date. Requirements:

- Python 3 with FastAPI, SQLite via sqlite3, Jinja2 for server-rendered HTML, and local disk for
  uploads. Single process, no Docker, no signup flow, no tenants. It serves one site: mine.
- Collection page: a shareable link taking a star rating, a written quote, and optionally a video
  or audio take recorded in the browser. Ask for name, role, company and a photo, let me mark which
  of those are required, and stop recordings at three minutes.
- Re-encode recordings to MP4 with a thumbnail in a worker off the request path, keeping the
  original. If ffmpeg is absent, store the upload and flag it rather than failing the submission.
- Moderation screen at /inbox behind an .env password: newest first, approve or bin, plus free-text
  labels I can attach to a review.
- Now the part I actually care about. A display is either a hand-picked list or a saved filter. A
  filter can require a minimum rating, a media type, a photo on the reviewer, a job title, a named
  company, one of my labels, a phrase that must appear, and a phrase that must not.
- When a review is approved, every filter it satisfies picks it up immediately, with no republish
  step. Each display shows whether it is filtered or hand-picked and how many reviews it resolves to.
- Let me override a filtered display by pinning up to three reviews to the front of it.
- Embedding: a one-line script tag plus an empty div. Attach a shadow root and inject markup the
  server already rendered, so the host theme cannot restyle it and the quotes sit in the response.
- Three presentations: a masonry grid, a swipeable row, and a slow vertical ticker. Theme, accent
  and radius come off data- attributes.
- A standalone wall at /wall with my logo, a call to action, and Review plus AggregateRating
  structured data emitted from the same query that renders the cards.
- A score badge: rounded average and total count, recalculated whenever a review is approved or removed.
- Public endpoints get a per-IP throttle and an invisible decoy field, since anything with a form
  attracts bots.
- README covering the embed snippet, the .env keys, and running it behind nginx with a TLS cert.
- Not building: fetching reviews I already have on Google, Trustpilot, G2, Yelp or an app store, or
  polling them for new ones. That is a parser per platform plus someone maintaining them, which is
  not a weekend. Teammates and outbound review-request email are out too.