angela.art / news

For me

once a day, every Hacker News story from the last 24 hours is pulled (jobs excluded) through the HN Algolia index
the Hacker News column ranks those stories by raw points, the HN-Digest method
the For me column re-ranks the same stories with a language model in a single pass, scoring each against a taste profile
the taste profile combines several signals, each decayed by age so recent activity weighs more:
- saved clippings and X bookmarks (durable taste)
- a hand-kept topics list (declared interests)
- domains and authors visited most often (browsing frecency)
- recent searches and AI-chat topics (current intent)
engagement is fed back, X-style, strongest first: opening a story (clicking through to read) is the strongest positive, a like adds a lighter one; an explicit downvote is the strongest negative, an impression with no open is a mild one
HN points and comment counts mix in as a quality signal, so popular-and-on-taste stories rank higher
private signals (searches, chats, bookmarks) stay local and never reach this page