Autonomous AI News Editor

A fully autonomous news editor that scrapes articles from trusted sources, translates them to Telugu, and publishes them to a CMS - all in a single browser-based process.

Quick Start

1. Install Dependencies

# Install Python packages
pip install -r requirements.txt

# Install Playwright browser (use python -m playwright, not just playwright)
python -m playwright install chromium

2. Setup Environment Variables

Create a .env file in the project root:

# CMS Credentials (REQUIRED)
CMS_URL=https://your-cms-url.com
CMS_EMAIL=your-email@example.com
CMS_PASSWORD=your-password
CMS_ROLE=State Sub Editor

# OpenAI API Key (REQUIRED)
OPENAI_API_KEY=sk-your-openai-api-key-here

# Browser Settings (OPTIONAL)
HEADLESS=false
SLOW_MO=400
USER_DATA_DIR=.playwright
SCREENSHOTS_DIR=artifacts/screenshots
DOWNLOADS_DIR=artifacts/downloads

Important:

• Copy env.example to .env and fill in your actual values
• CMS_URL, CMS_EMAIL, CMS_PASSWORD, and OPENAI_API_KEY are REQUIRED
• HEADLESS=false means browser will be visible (recommended for first run)
• SLOW_MO=400 adds 400ms delay between actions (makes it easier to watch)

3. Run the Editor

python main.py

What Happens When You Run

1. Browser Opens (visible, headful mode)
2. CMS Login - Automatically logs into your CMS
3. Article Scraping - Tries to scrape from BBC/Al Jazeera/NBC (in order)
4. Summarization - Generates English summary (300-360 chars)
5. Telugu Translation - Translates to Telugu (300-360 chars)
6. Category Selection - Decides category using keywords
7. CMS Publishing - Fills form and publishes article
8. Success - Stops when one article is published

The system will try up to 5 articles until one is successfully published.

Architecture

Key Principle: The Browser Agent is the ONLY decision maker. No external schedulers, databases, or API servers.

Components

• main.py: Controller loop (tries up to 5 articles until one is published)
• agent/browser_agent.py: Playwright browser management
• agent/scraper.py: News article scraping from BBC/Al Jazeera/NBC
• agent/summarizer.py: English summarization (300-360 chars)
• agent/telugu_writer.py: Telugu translation with fallbacks
• agent/category_decider.py: Simple keyword-based category selection
• agent/cms_publisher.py: CMS login, form filling, and publishing

Features

• No Database: All in-memory, no persistence
• No Scheduler: Single run, publishes one article
• No Hard Blockers: Quality checks warn but never block
• Visible Browser: See exactly what's happening
• Automatic Fallbacks: Telugu failures → English, publish failures → try next article

Category Rules

• Spiritual: temple, god, puja, astrology
• Sports: cricket, match, player, ipl
• Entertainment: actor, film, cinema, movie
• Health: health, diet, disease, medical
• International: Only if clearly foreign (USA, China, etc.)
• National: Default for everything else

Telugu Style

Matches TV9/Eenadu/Sakshi digital news style:

• Natural Telugu (not literal translation)
• Newsroom vocabulary
• Short, punchy sentences
• 300-360 characters for summary

Troubleshooting

"OPENAI_API_KEY not found"

• Make sure you created .env file (not env.example)
• Check that OPENAI_API_KEY=sk-... is in .env

"CMS credentials not found"

• Make sure CMS_URL, CMS_EMAIL, and CMS_PASSWORD are in .env

Browser doesn't open

• Make sure Playwright is installed: playwright install chromium
• Check that HEADLESS=false in .env

"No articles found"

• Check internet connection
• News sources (BBC/Al Jazeera/NBC) might be blocked in your region
• Try running again (different articles may be available)

CMS login fails

• Verify CMS credentials in .env
• Check that CMS_ROLE matches exactly (case-sensitive)
• Make sure CMS URL is correct

Article scraping fails

• Network issues
• News source website structure changed
• Try running again (will try different sources)

Logs

Check artifacts/logs/automation.log for detailed logs of what happened.

Screenshots

Screenshots are saved to artifacts/screenshots/ if errors occur.