Co-founder, Omber

I was the hiring manager. Building the team at WAND meant posting a role and getting back a few hundred résumés, all claiming the same things in the same words in the same order.

A résumé is a document about a person. What I needed was the person.

And I was a hands-on director, so reading the stack was work I did after the work. Then every name I pulled out of it meant a thirty-minute intro call to learn something I could have known in three. I did not have time to do those calls properly, which is a polite way of saying I did some of them badly — and the cost of that lands on the candidate, not on me.

Have the candidate record a short video and send it with the résumé. Then there is no intro call, because you have already met them.

I found a co-founder and we started building. This was outside a full-time job, so I recruited two engineers in Ukraine — one backend, one frontend — and ran the pod around my own hours. That constraint shaped the whole thing: nothing could depend on me being awake, so the work had to be specified well enough to be picked up in a different timezone and finished without me.

We got it in front of real hiring teams. One of them was a Fortune 500 software company.

They told us the same thing often enough that it stopped being feedback and became the roadmap. The video intro was useful. The thing they would pay for was assessment — not just hearing the candidate, but knowing whether they could do the work.

So I designed one. Hiring managers build the assessment with AI rather than from a blank page, and AI helps score the answers that come back.

Handing an LLM a free-text answer and asking whether it is any good gets you a number with no account of itself, and not the same number twice. Pointed at people applying for jobs, that is not a rough edge. That is the product doing harm.

So an assessment is not finished until its rubric is. Every question carries an explicit structure of what a good answer contains, the model scores against that structure rather than against its own taste, and the hiring manager can see which part of the rubric produced which part of the score.

That was built for equity. It also turned out to be the fastest way to score, which is the usual shape of that trade — the thing you do to be fair to the candidate is the same thing that gives the hiring manager a number they can defend.

Omber is the second time I built the thing I was standing in the middle of needing. The first was a bus app, built because I kept missing the bus.

That is the whole method, run on myself. Sit with the person at the table, describe their better day, then work backwards to the stack. It is easier to trust when the person at the table is someone else's customer, but it started here, with my own hiring problem and four hundred documents I could not read.