I've been a daily user here for almost 15 years. Over that time, my interests have shifted. I find I'm now more interested in deep-dive technical posts and personal blogs than the big tech announcements that often dominate the front page. The "top" feed was starting to feel stale, and I was spending more time digging through "new".

So, I decided to build what I wanted: a personalized "For You" feed.

Link: https://hn.shaped.ai

It's a simple concept: you log in with your normal HN credentials, and as you favorite stories, it learns what you're interested in and re-ranks the feed to show you more of that content.

How it was built (a 2-day hackathon):

The Client: I used an AI coding assistant (lovable.dev) to generate the initial React/Next.js client. It was surprisingly effective at getting a functional baseline up and running quickly. The Backend: Since HN's official API is read-only, I set up a lightweight Supabase backend. It uses edge functions to proxy login/voting requests to HN's unofficial API and a Postgres DB to cache posts and user events (favorites, etc.). The Personalization: The ranking is powered by my own company's platform, Shaped. It ingests the posts and your favorite events in real-time.

The core of the ranking logic is a configurable formula. It's essentially the classic HN algorithm with a personalization term multiplied in:

(item.score / score_penalty + content_similarity) / (time_decay)

The content_similarity is calculated by comparing a post's text embedding to an embedding of your recent favorites. The best part is that you can actually play with the score_penalty in the UI to make the personalization stronger or weaker.

This is very much a v1. I'd love to get your feedback. Does the personalization feel right? Any bugs? What's missing that would make you use it daily?

Next on my list are things like collaborative filtering (once there's enough data!), semantic search, and a "similar stories" feature. Here's also a more detailed write-up about how it was built: https://www.shaped.ai/blog/building-a-hackernews-for-you-fee...

Thanks for checking it out!