Systems

Two systems, described exactly as they are.

One is live and does real work every day. The other is in active development. Status labels on this page are literal.

  • AI Control Center Live
  • Rahyab In development
AI Control Center

Our own AI publishing system.

Live  It plans, drafts, checks and publishes content for the channels we run. It is an internal system — not a product we sell today.

  1. InputA topic, or news found with cited sources.
  2. AI draftAI writes the post, script, image or video.
  3. Automated reviewNews posts are checked against the sources they cite; content rules are enforced in code.
  4. Human decisionFlagged items are held in Telegram: publish anyway, or discard.
  5. Publish & logPosts that pass go out; publishing is recorded in an audit log.

What happens on each channel

  • Facebook news posts are researched with cited sources, then checked by a separate automated review that reads those sources. Posts that pass publish automatically. Posts that fail are not published: they are held and sent to the owner in Telegram with the reasons, to publish anyway or discard.
  • YouTube videos are always uploaded as private. The owner reviews them and makes them public himself.
  • Instagram receives image posts with a caption.
  • Facebook and Instagram never receive video. This is enforced in code.

Controls we rely on

  • A repeat-subject check, so posts don't keep returning to the same story.
  • A single pause switch that stops all publishing.
  • Accounts connected through official OAuth; sign-in tokens stored encrypted.
  • For YouTube, only upload and read-only permissions.
  • An audit log of connections, approvals and publishing.

AI assists. Humans stay in control of what is flagged — and of what goes public on YouTube. Not every post is approved by a person before it is published; that is why the rules above exist.

Rahyab

Local business intelligence that shows its sources.

In development  Rahyab runs on our own server for internal testing and is being actively developed. It has no outside users and no public sign-up yet.

source said

What a listing, a seller or a channel post claimed. Anything AI extracts is stored here, with capped confidence — it can never become a verified fact on its own.

rahyab verified

What a human admin actually checked. Never set by AI or by a parser.

rahyab calculated

What Rahyab computed from stored evidence, such as medians, best rates or scores.

What is built

  • An observation history: new information is added next to the old, never written over it.
  • A first real-estate model: properties separate from the listings that mention them, local area units normalized to square metres, and a review queue for likely duplicates.
  • Background workers that retry failed work and flag anything that keeps failing for manual review.
  • Sign-in through Telegram, and a Telegram Mini App shell.

Built-in care

  • Public search never shows full phone numbers; revealing a contact needs a login and is logged.
  • Duplicates are never merged on a phone number alone.
  • Request rate limits on reads, writes and logins.
  • Public or explicitly authorized data only.
  • FastAPI
  • PostgreSQL + PostGIS
  • Redis
  • Background workers

Need a system like these?

Tell us what you want to build. The Start a Project page prepares an email for you to send.

Start a Project