Overview
Basement Cowboy is a complete human-in-the-loop AI publishing pipeline. It continuously scrapes more than 180 news sources, scores every article with a custom 7-factor ranking algorithm, and surfaces the best candidates in a curation dashboard. Approved stories get an AI treatment — summarization, categorization, and a generated feature image — before publishing to WordPress fully SEO-optimized.
The Pipeline
Scraping runs on BeautifulSoup with Playwright-driven Chromium for JavaScript-heavy and bot-protected sources. The ranking layer weighs seven signals to sort thousands of daily articles down to a manageable shortlist. From the Flask dashboard, an editor reviews candidates, triggers AI enhancement, and publishes — the system handles WordPress REST and GraphQL delivery, categories, and metadata.
Cost-Aware AI
The AI layer is deliberately pragmatic: GPT handles summarization and categorization, DALL·E 3 generates feature imagery, and the UI shows the estimated cost of every enhancement before you commit — a small design choice that makes the economics of AI content workflows visible rather than surprising.
Engineering Highlights
- Dockerized with Gunicorn and docker-compose; GitHub Actions CI/CD
- File-based JSON storage — zero-database deployment for simple hosting
- Full lint/type/security tooling: pytest, black, mypy, bandit
Outcome
A shipped v1.0 that automates the grind of news curation while keeping editorial judgment — and cost control — in human hands.