Hey PH! I'm the maker of myown.news. The idea started from a simple frustration: every news product is a feed. Feeds are infinite, algorithm-driven, and all look the same. But newspapers had something right — a finite, curated edition with editorial structure. I wanted to see if AI could replicate that editorial process, not just summarize links. Here's what's under the hood: AI newsroom pipeline: A scheduler crawls HN, arXiv, GitHub Trending, RSS feeds, and newsletters every few hours. Then an LLM (Claude Opus) acts as editor-in-chief — it cross-references sources, deduplicates across topics and past editions, selects stories by news value, and writes a full edition with proper article types (lead, feature, brief, filler) and editorial commentary. Not a wrapper: The output is structured Markdown with semantic metadata (importance, source type, figures, quotes). The frontend parses this into a real multi-column broadsheet layout — no templates, no cards. Edge-native stack: Cloudflare Workers + D1 + R2 + Hono. The entire app runs at the edge. Image quality is scored by GPT-4 Vision with a 4-level fallback chain. Parallel LLM translation: 8 languages generated concurrently using a mix of Azure, Claude CLI, and Subrouter to optimize cost and latency. Cross-edition dedup: A similarity pipeline prevents the same story from appearing across days or topics. Everything is built by one person. I'd love technical feedback — especially on the editorial pipeline and layout rendering. What would you want to see next?

MyOwn.News

Your world, typeset daily.

Your world, typeset daily.