Knight-Panther/Telo-watch-tower — reverse-engineered prompt

Build me an AI powered media monitoring app called Watch Tower. I want it to pull articles from lots of RSS feeds, save them, catch exact and near duplicate stories, then have AI score each article from low importance to high importance and write a short summary. High scoring stories should move forward automatically, low scoring ones should be rejected, and the middle ones should wait for manual approval in a web dashboard.

Once something is approved, it should be able to translate the title and summary into Georgian, create a clean news card image, and publish the post to Telegram, Facebook, and LinkedIn. I also want keyword based Telegram alerts, a daily digest with draft approval, and a dashboard where I can manage feeds, sectors, scoring rules, posting language, prompts, analytics, and pipeline status in real time.

Please make it production ready, secure, and easy to run with Docker, with sensible environment setup and restartable worker stages. If anything is unclear, look up current docs online and fill in the gaps.