Daily AI Brief - Every morning at 6 AM, AI news shared like a newspaper
Translated by AIView original
Screenshots
Screenshot 1 of 3
AI news comes out every day, and it was hard for me to go through overseas blogs, papers, and communities one by one. So I built an AI news briefing you can skim each morning like a newspaper.
How it works
-
Every day at 6 AM, it automatically collects from 47 sources at home and abroad: the official blogs of OpenAI, Anthropic and Google DeepMind, TechCrunch and The Verge, arXiv and HuggingFace papers, Hacker News and GeekNews, and AI Times and Electronic Times.
-
English articles also come with a Korean title and a 2-3 line summary, so you can see the key points before opening the original.
-
Each article gets a field (language models, agents, dev tools, MLOps, research, industry policy), tags, and an importance level.
-
Important articles run as the front-page lead, and articles with hot community reactions get moved up to "Trending now."
-
The design takes cues from paper newspapers. I used a masthead, a serif nameplate, and thick rules to divide the page, so it feels more like reading today's edition than flipping through a feed.
-
Dark mode and mobile are supported too.
No login or subscription needed. Just open the link and you can read right away.
How it was made
I built it together with Claude Code.
We wrote a PRD together, split it into step-by-step work plans, and implemented one step at a time.
The goal was "running costs close to 0 won."
-
With no server, a GitHub Actions cron runs the collection script once a day.
-
The results are stored in a single SQLite file (data/app.db), which is committed to the repo as is.
-
When building, Next.js 15 reads the DB and generates static pages, which are deployed to GitHub Pages. So there's no DB server and no hosting cost.
-
The thing I cared about most was LLM token cost.
-
URL normalization and hashing filter out duplicates first, and only new articles not in the DB get sent to the LLM.
-
For each article, Claude Haiku is called exactly once, and it returns the Korean title, summary, field, tags, and importance together as JSON.
-
I set the design around a newspaper concept called 'Edition,' and collected the color, typeface, and spacing rules in DESIGN.md so I could refer to them every time I worked on the UI.
- Built with
- Claude Code
- Other
Comments 0
No comments yet
Be the first to comment.