What it does This project scrapes Hacker News, pulls the main content from each linked article using a machine-learning-based extraction algorithm, generates summaries via OpenAI’s GPT-3.5-turbo API, and renders everything as a static site deployed to GitHub Pages. If OpenAI is unreachable, it falls back to a local Google T5 model. The result is a readable digest with thumbnails, embedded videos/PDFs, and RSS feeds.

The interesting bit The whole pipeline runs on GitHub Actions and publishes to GitHub Pages — no server, no database, just scheduled scraping and templated HTML. The content extraction uses a scoring algorithm rather than simple heuristics, and the project even localizes summaries into Chinese with a single prompt tweak.

Key highlights

• Static site generated entirely via GitHub Actions (badge shows it’s actively building)
• Dual-model summarization: OpenAI API with local Google T5 fallback
• Auto-extracted illustrations and embedded media (YouTube, PDFs, GitHub gists)
• Sortable and filterable by points, comments, or time; RSS fully supported
• Chinese translation available via LLM prompt modification

Caveats

• The README’s “score algorithm” for content extraction links to a Jupyter notebook tutorial, but the actual implementation details aren’t shown inline
• TODO list includes switching to the official Hacker News API and improving web scraping (currently not using headless browsers)
• “Seamless” appears twice in the README; the project itself is more pragmatic than that word suggests

Verdict Worth a look if you want a self-hosted HN digest or need a reference for building cheap, serverless content pipelines. Skip it if you need real-time comments or a polished UI — the author admits the homepage could be prettier.