I burned 100s of hours into my 400 level classes, some of those hours were completely wasted on things that I could never figure out how to get them to work.
The matching algo is very raw at this point, so if you find posts that _should_ have matched but didn't, please reach out (email is in my profile) and I'll see what I can do.
> But the optimal for those isn't to post a generic who wants to be hired but have a LLM rewrite your resume to match the job description isn't it?
If anyone was depending on this to actually make interview decisions, yes, that would be something a candidate might do. Fortunately there's not that pressure on this project. :D
I left an incomplete section in the post body, then went shopping.
So here it is:
---
Some observations when looking at the matches pair up right now:
- individuals looking for jobs have extremely general postings
- companies looking to hire have extremely specific postings
- generally, match scores are very low (<50%) because of this mismatch
- individuals don't typically mention domain knowledge in any specific way
- individuals don't typically mention prior work history beyond time-in-industry (this makes sense on a public forum)
- individuals typically list technologies used
I don't have any conclusions to make right now, no advice for job posters or job seekers. It'll take two or three more of these before I've seen enough to start gelling the domain problems in my head.
I wouldn't bother unless you just want to try to compile stats. The active job seekers and the open positions from more than a month ago probably aren't valid today.
It's a bit like backtesting, in finance. Quite useful to establish whether your algorithm is sane and coherent, no use at all in determining if it is actually useful, to my understanding.
There's a few metrics: seniority, salary range, enumerated skills (as requirements or as a person's skill set), and domain specific knowledge.
These are far from perfect, and I think in the future I will probably tweak how matches are scored, because how job postings are written and how 'Who Wants to Be Hired?" posts are at odds.
The former is one-to-many, where the job posting enumerates very specific requirements in order to filter that 'many' down to as few eligible candidates as possible.
The latter is one-to-many, where the 'I'm for hire' posting seeks to match as many of the 'many's as possible. This means down playing domain experience or specific technologies in order to appeal to as many possible viewers as possible.
Trying to pair 'precise and specific' with 'as general as humanly possible' is ... hard. Here we use 'tags' extracted from postings to match the domains and skills together. It isn't perfect. From what I've seen in this first run there aren't any matches that make it over 50% -- where 50% is a heuristic for the metrics mentioned above. This just means there's room for improvement.
The other side is the old game of 'tailor your resume to the job', which isn't in the spirit of 'Who Wants to Be Hired?' posts. You could very easily game this by just dumping a bunch of keywords into your post.
Also always accompanied by fear.
reply