A Daily Mandarin Newscaster Built on Cloudflare Workers

What it does Every day, a Cloudflare Worker scrapes the top stories from Hacker News, uses an LLM to draft Chinese summaries and full broadcast scripts, then runs text-to-speech to generate a Mandarin audio episode. The output is served through a web player and a standard RSS feed syndicated to Apple Podcasts, Spotify, YouTube, and 小宇宙.

The interesting bit The project treats Cloudflare’s edge stack as a complete content factory: Workers run the crawl and AI generation, R2 stores audio files, KV holds metadata, and the audio merging step even relies on Cloudflare’s browser rendering service. It is essentially a serverless broadcast studio.

Key highlights

• End-to-end automated pipeline from HN crawl to published podcast, running daily
• Generates both Chinese audio and readable transcripts for every episode
• Web interface built with vinext (Vite + React Server Components) and deployed to Workers
• Syndicated across major podcast platforms including RSS, Apple Podcasts, Spotify, and YouTube
• Uses OpenAI API for content generation and TTS for voice synthesis

Caveats

• Local development is bumpy: the README notes that Edge TTS can hang during local runs, so you may need to comment out audio generation to debug other parts
• Audio merging depends on Cloudflare browser rendering, which does not work in local development and requires remote debugging

Verdict A solid reference if you are building automated content pipelines or curious about running LLM-plus-TTS workflows on Cloudflare’s edge. Not for you if you need a local-first podcast tool or a generic HN reader.