루리의 줍줍 - 1,360-page Korean info site built by AI, $0 server cost

I'm not a full-time developer. I describe what I want in plain Korean and an AI coding agent (Claude Code) does the planning, coding, deployment, debugging, images and short videos. 2 days, 110 commits, ~1,360 static pages that rebuild every hour, server/API bill $0 (domain and the AI subscription are separate). How the $0 works: GitHub Pages for static HTML; Cloudflare Workers + D1 free tier for login, bookmarks, visit counter and web-push subscriptions; a public vision model on Workers AI (capped per day) for a fridge-photo-to-ingredients tool; public RSS (only where robots.txt allows), Upbit's public crypto API and Google Trends RSS as data; Windows Task Scheduler runs collect > process > build > git push every hour. Technical bits I found interesting: grouping the same story across 15 news outlets by comparing against a seed article (chain-linking merged unrelated stories); a "verification signals" score that checks whether numbers and double-quoted quotes in a headline appear in the body; analyst reports modeled as nodes (firm, direction, old to new price target) rolled up per stock; search cleanup with content-hash sitemap lastmod and pre-rendering JS-drawn screens to static HTML (text visible to crawlers went from 687 to 13,605 characters). Where the AI got it wrong: an undated year-old RSS item showed up as "today's report" (fixed with RSS > URL > page-metadata date resolution plus a 7-day cutoff), single quotes used for emphasis were treated as quotations, and generated characters had odd hands (only hand-checked images are used). The site is in Korean — use the 🌐 button (top right) for English. Build log: https://sungprofile.cloud/ailog/...