I’m building Hiredra as a personal job-search system. It combines direct job discovery, LLM-assisted resume preparation, code-based scoring, and human review; I have not completed end-to-end testing yet.
My firm Hiredra constraint is: no LLM scores a job.
I believe LinkedIn applications need to be submitted in the first hour.
I describe the potential LinkedIn queue as 100, 200, 1,000 people ahead of an applicant.
I crawl ATS boards directly to find the jobs people aren't even seeing.
In Hiredra, the LLM extracts skills from the JD and generates the custom resume.
I build the score with code: knowledge graph, vectors, similarity search.
I consider an LLM’s 70 or 80% score dicey when two jobs are really close.
One close-match example is a job that needs skills very close to what I know but that I don't technically have.
Another close-match example is a job that has 50-60% of my exact skills.
My assessment of the close-match ranking problem is: An LLM can't rank those properly.
In Hiredra, a human still reviews every application.
I estimate that human review takes About 10 seconds each.
I keep human review because If one mistake is in all of them, you'd never know otherwise.
My Hiredra status is not fully tested end to end; interview numbers unavailable.
I’ve asked people who hire or build in this space whether they would trust a code-built score over an LLM-built one, and where my approach may be wrong.