Case study
Simorgh News: an AI newsroom for a Telegram channel
How one person runs a news channel with an AI pipeline, a human review step and a server that never sleeps.
- Python
- FastAPI
- aiogram
- Next.js
- Docker
- Azure
The problem
Running a news channel alone means reading dozens of sources and rewriting posts every day. It eats hours, and posting stops whenever you are busy.
How it works
Collect
RSS feeds from BBC and Al Jazeera are collected every 30 minutes into a database.
Write
An LLM rewrites an English article into a Russian post in the channel's voice. If a free model is overloaded, the next one in the chain takes over.
Review
Drafts wait in a web panel and in a Telegram bot. The editor approves, edits or rejects them.
Publish
Approved posts go to the channel in one click.
Architecture
Decisions that mattered
A human stays in the loop: nothing is published without approval.- Answers that are not in Russian are rejected automatically, because models sometimes leak their reasoning in English.
- Any OpenAI-compatible provider works through configuration, so changing providers needs no code changes.
- One small cloud server runs the API, the bot, the panel and HTTPS behind Caddy, with Docker Compose.
Result
Running 24/7 since September 2026: 632 articles collected and 8 posts published. Publishing a post takes one click.
What I learned
Free AI services change often, so the system must not depend on any single one. Real numbers and a simple review step build more trust than any amount of automation.
Want something like this?
Tell me what you do by hand today, and I will suggest how a bot can take it over.