An increasingly good question to ask oneself these days: how could this be done without direct usage of LLMs / without ML even? Then as a second step ask yourself how potential assistance from generative tools and perhaps ML could improve those solutions.
I made a version of this using boring tech (Postgres, Django, Python) by just counting the number of times a book link (not just books, youtube videos, arxiv papers etc) is posted in Hacker News comments. I also did a bunch of calculations around the poster of the link and replies to the posted link. The reality is that boring tech does not get attention and engagement.
Happy to hear (with Postgres or Python or Django) how you are going to filter out irrelevant comments, from comments mentioning books. And then how you're going to extract the book title and its author from a comment?
That doesn't look trivial at all to me... but... if you did that, please share with us a Github repo.
The site is http://hnlikes.com/ and on the About page I describe how it was done.
Like I mentioned it only looks for books that are linked (or videos, arxiv papers, Wikipedia etc.). I then use the link to get information from the site itself.
I calculate scores for the given link weighted by a number of metrics.
I'm sorry if I gave the impression this was fancier than it actually is.
I think a lot of the basic tasks with LLMs can be done without ML but what I find interesting is learning how to use LLMs to do interesting things and the ceiling being a lot higher than with traditional methods.
Learning how to use a new tool with older type of work can be useful and enlightening.
