The basic process is fetching https://hnrss.org/newest then piping each item through gpt-4.1-mini with just the URL and title (I think fetching each actual page is overkill and a bit greedy) and asking it to figure out if something is likely to be development related. If it's a confident yes, that item then gets into the final feed. It cuts the firehose down quite a bit just with that simple approach.
The basic process is fetching https://hnrss.org/newest then piping each item through gpt-4.1-mini with just the URL and title (I think fetching each actual page is overkill and a bit greedy) and asking it to figure out if something is likely to be development related. If it's a confident yes, that item then gets into the final feed. It cuts the firehose down quite a bit just with that simple approach.