AIHawk
A job-application bot that has Jev score each posting for fit before it applies.
A job-application bot that has Jev score each posting for fit before it applies.
One state holds the candidate's CV and the postings keyed p00, p01 and so on, each reduced to title, company, location and up to 24,000 characters of description. Every posting gets its own Noul question in the same request, asking whether the job is the same kind of work at the candidate's level, matches what the CV shows, and is near their places or remote for them. The yes probability becomes a score from 1 to 5 in the local store, and the app works through postings best first.
For posting p00, is this the same kind of work at the candidate's level, matching what their CV shows, and based near their places or remote for them?
The state carries the candidate's CV and the postings keyed p00, p01 and so on, each as title, company, location and the start of the description. One Noul question per posting is sent in the same request.
The app reads postings already found on this computer from its local library and adds the new ones to a store with the status new.
Jev scores the next postings the search ranked best, with the CV as state and one fit question per posting, batched up to 512 per request.
A request the endpoint calls too large is split in half and retried; a posting that got no verdict is marked passed over instead of being asked again.
Each yes probability is stored as a score from 1 for a sure no to 5 for a sure yes.
The screen lists scored postings best first, and the apply step fills the application form in a browser with the person's CV and profile.
It asks for a calibrated probability per posting instead of a written ranking, batches up to 512 postings into one request, and sets aside a posting that got no verdict instead of retrying it.