Bakhrushin Museum News — Telegram automation

Automated system for collecting theatre-related news from the internet, generating Telegram-ready summaries with AI (OpenAI-compatible provider polza.ai), moderating, and publishing to the Telegram channel @bakhrushinmuseum_news.

What it does

• Collects news from RSS and HTML sources
• Stores data in PostgreSQL (with Alembic migrations)
• Generates summaries via polza.ai (OpenAI-compatible)
• Provides a web moderation panel (FastAPI + Jinja2)
• Publishes to Telegram via aiogram
• Runs background jobs via Celery + Redis + Celery Beat

Pipeline:

Requirements

• Docker + Docker Compose (Compose v2)
• Ubuntu 22.04 / 24.04 recommended

Quick start

1) Configure environment

Fill at least:

• TELEGRAM_BOT_TOKEN
• TELEGRAM_CHANNEL
  (default:
  @bakhrushinmuseum_news
  )
• POLZA_API_KEY

2) Run services

3) Initialize database

4) Seed initial sources

5) Open web panel

• Posts moderation: http://localhost:8000/
• Sources management: http://localhost:8000/sources

Telegram setup

1. Create a bot via @BotFather and set
   TELEGRAM_BOT_TOKEN
   .
2. Add the bot as admin to the channel @bakhrushinmuseum_news with posting rights.
3. Set
   TELEGRAM_CHANNEL=@bakhrushinmuseum_news
   .

Moderation & publishing

Statuses:

• pending
  — waiting for moderation
• approved
  — publish ASAP (publisher will pick it up)
• scheduled
  — publish at
  scheduled_at
• posted
  — published successfully
• failed
  — publishing failed (see
  editor_notes
  )

Approve publishes immediately

• Open a post → optionally edit text → Approve

Schedule uses Moscow time (MSK)

• Open a post → set
  scheduled_at
  in MSK format
  YYYY-MM-DD HH:MM
  → Approve + Schedule
• DB stores schedule in UTC; UI accepts and displays MSK.

Sources management

Open: http://localhost:8000/sources

You can:

• add new sources (type:
  rss
  /
  html
  )
• edit
  parser_config
  (JSON)
• enable/disable sources
• delete sources
• Test a source URL (stores
  last_status_code
  ,
  last_checked_at
  ,
  last_error
  )

When a source returns 401/403

Some sites block bots. Use the Test button and disable such sources, or switch to a different URL/feed.

Cutoff filters (avoid old news)

To prevent collecting or AI-processing old items, use sliding window cutoffs:

Rules:

• NEWS_NOT_BEFORE_DAYS
  : collector skips items older than N days if the item date is known.
• AI_NOT_BEFORE_DAYS
  : AI step skips items older than N days if the item date is known.
• Items without a date are not blocked by cutoffs.

Collector and AI steps log skipped items as:

• [cutoff] skipped by NEWS_NOT_BEFORE ...
• [cutoff] skipped by AI_NOT_BEFORE ...

Operations

Check services

Logs

Run jobs manually

Backup

Dump database

Restore

Troubleshooting

"Received unregistered task" in celery_worker

This means the worker did not import tasks. In this project it is fixed by importing

in

app/workers/celery_app.py

.

Restart:

Port 6379 already in use

Redis port conflict on host. Either stop local Redis, or remove port mapping for Redis in

.

Alembic errors about sqlalchemy.url

This project uses

to load

DATABASE_URL

from

.env

and sets

configuration["sqlalchemy.url"]

at runtime.

Project structure

• app/api/
  — web routes
• app/templates/
  — Jinja2 templates
• app/collectors/
  — RSS/HTML collection
• app/parsers/
  — text extraction
• app/ai/
  — AI summarization
• app/workers/
  — Celery tasks
• app/tg/
  — Telegram sender
• migrations/
  — Alembic migrations