How this got built
neta was not designed. It accreted. Each version existed because the one before it hit a wall.
It started as one script. I was applying for jobs and I was tired of typing the same answers into the same forms — name, work authorization, desired salary, how did you hear about us. So I wrote something to fill one board’s form for me. That was the whole idea.
One script could not remember what it had already done, so I gave it a database. Now it knew which postings it had seen and which it had applied to.
Pointing it at each posting by hand did not scale, so it started downloading the postings itself — reading the open roles off the job boards instead of waiting to be told where to look.
Then it had a bad habit: it applied to everything and scored the fit afterward. That is backwards — applying first and judging later is how you send an application to a job you were never right for. So I turned it around: score first, apply only to what fits. That one change is why the system says no far more than it says yes. The filtering turned out to be most of the value.
Scoring by keyword rules was not good enough — a résumé is not a bag of words. So a language model started reading each posting against my actual background and scoring the match. A local model does the volume; a cloud model handles the hard calls. Cheap first, expensive only when it earns it.
The forms kept asking things the system had never seen. Instead of guessing — a wrong answer on a real application is worse than no application — it started logging every question it could not answer, learning it, and reusing the answer next time. It got better the longer it ran.
The boards are not alike — Greenhouse, Lever, Ashby and Workday each do it differently — so each got its own handling.
None of that was planned up front. It grew the way real tools grow: one person solving their own problem, one wall at a time.
That is why neta is sold as a working example rather than a course. It is not a tidy teaching exercise — it is the actual thing that grew, with the messy history, the real decisions, and the data to show for it: 11,700+ real job applications, and every number has its query.
If you can run Python, you can own the whole history and take it apart.
