Hi HN,
TL;DR: I made a website that takes all the top stories on HN and categorizes them into one of ten topics of interest using LangChain and GPT-4.
I've always liked the idea of getting a personal mix of news tailored to my interests. Hacker News is pretty close to that ideal, which is reflected in the time I spend on here every day.
It's a great mix, but playing with LangChain, I got the idea for a weekend project. Roughly speaking most posts on HN fall into one of these categories:
* Programming, Software & Computer Science
* AI, Data Science & Analytics
* Business & Entrepreneurship
* Science & Research
* Cybersecurity & Digital Safety
* Design, User Experience & Creativity
* Finance & Economics in Tech
* Work Culture & Career Development
* Media, Content & Communication
* General & Diverse Interests
Using langchain with GPT-4 Turbo (JSON mode), I sort every top submission into one of the above. Costs for using the API are currently at 15$ a month, with few optimizations.
For ease of use I also added the possibility to consume the news topics via a responsive webpage, RSS and email digest (daily,weekly,monthly).
A few minor comments.
1/ Current homepage is a bit overwhelmed to read.
2/ The API cost can be greatly reduced (eventually to zero) by caching results, and some text similarity tricks.
3/ Is this going to be a commercial product ?