Hiring process

WHY THIS PROCESS

AI cheating is a process problem, not an interview problem. You can’t out-question someone with a hidden assistant in their ear — you design around it. Watching you work in real time, real work products, multiple stages of human judgment.

Every stage tests a different thing. The build day tests how you actually work. The depth conversation tests whether you know what you claim to know. References tell us about how you fail and recover. None of these can be faked.

THE SIX STAGES

Built around live evidence.

01 / 07

01SAMPLE OF WORK

Send something you built.

Not a resume — a thing. A paper, a repo, a demo video, a system that runs. One piece representative of what you’d do here.

Async No AI question 5 min to submit

Why this works

Anyone can ChatGPT a resume. Nobody can ChatGPT a working system they’re about to defend.
The form of the artifact tells us how you think before we’ve said a word to you.
If it’s yours, you can defend it. If it isn’t, the next stage will surface that.

02FOUNDER CALL

Walk the founder through it.

30-minute video call with Romir. He asks why you built it that way, what the alternatives were, and what AI tools helped vs got in the way.

30 minutes Video call One founder

What we’re testing

Intimate familiarity with your own work. AI can’t fake the third or fourth follow-up.
How you frame trade-offs. Builders talk about what they’d do differently. Imposters defend.
Honest accounting of AI usage. What did it help with? Where did you have to override it?

“What’s the part of this you’re most embarrassed about?” is one of the questions we’ll ask.

03BUILD DAY

Spend a day building live.

Paid. Live and real-time — over video, or in person if we’re in the same city. Five to six hours. A real problem from the codebase. Your laptop, full internet, every AI tool you want. The founder or a senior IC sits with you the whole time.

5–6 hours Live Paid ($500–1k) AI tools encouraged

What we’re watching for

How you frame the problem. The first hour predicts the next five years.
When you reach for AI and when you don’t. The judgment is the assessment.
Whether you can spot bad AI output. Pattern-matching tools fail in ways the user has to catch.
How you debug. Most candidates lie about this. We watch it directly.
Whether the thing actually works at end of day.

This is the stage. If you’ve faked the previous two, this is where it surfaces.

04DEPTH CONVERSATION

Conversational depth check.

90 minutes, live — over video, or in person if we’re in the same city. Conversation, not a quiz. The founder and (when relevant) a team member who works in your area.

90 minutes Live Conversational

The test

Can you go three follow-ups deep without losing the thread? AI answers level-1; people who know the thing answer level-3.
Questions are drawn from your claimed expertise — not a stock list.
“Walk us through why this paper’s claim about cross-embodiment transfer doesn’t replicate.” Tests if you read papers or cite them.
“Where does this approach fall apart in a real fire scene?” Tests operator empathy.

05FOUNDER LUNCH

Time with the founder.

Just time together. No agenda. No deck. Over a live call, or over a meal if we’re in the same city. You learn how Romir thinks; he learns how you think. We don’t pretend this is anything other than what it is — we want to spend time with you.

~90 minutes Live Undivided attention

What we’re actually assessing

Cultural fit. Do you make the room sharper or duller?
Mission alignment. Are you here because of the operators, or because it’s a hot AI company?
Ego. The work is hard. Big egos break.
Taste. What do you admire? What annoys you? Why?

06REFERENCES

References, done seriously.

Three references. Two peers or managers, one person who worked for you or alongside you in a hard moment. Each call is 30 minutes. None of them are pro forma.

3 calls 30 min each Specific questions

The one question that matters

“When this person was wrong about something hard, how did they figure out they were wrong?”
AI doesn’t make humans introspective. References either confirm that quality or they don’t.
If a reference goes vague here, that’s the signal.

07FIELD DAY SHADOW

Sometimes: shadow a field day before signing.

For finalists, sometimes required. You join us for one of our quarterly field days — a ride-along with a fire crew, a range day, an EMS shift, a Stryker triage tour — before signing. We watch how you watch the operator.

1 day In the field with operators Travel + gear covered

What this last step tests

Operator empathy. Can you see the work through the operator’s eyes? Or do you reduce them to a problem statement?
How you react under conditions you don’t control. A fire scene at 3am is the inverse of an office. The difference matters.
Do you come back with a problem worth solving? Every field day generates one for the roadmap. We want to see yours.
Not every role requires this last step. Senior research and operator-facing roles usually do.

PRINCIPLES

How we promise to run the process.

01

Live is the test. Every stage happens in real time — over video, or in person when we’re in the same city. For final stages we’ll fly you out to meet the team; travel is paid up front.

02

Build day is paid. $500–1,000 depending on level. Your time matters whether or not we hire you.

03

AI tools are allowed throughout. We assess your judgment about AI, not your ability to type without it.

04

You meet humans, not panels. Every interview is conversational and one-on-one or two-on-one.

05

We tell you what to expect. No surprise stages. No surprise topics. We share the rubric.

06

2-week response window. No black-hole rejection. If we say no, you get a reason.

Read the process. Apply if it sounds right.

Open roles are on the careers page. Pick one, apply, and we’ll be in touch within two weeks.

Open roles →