Product Builder, Vault

You've been shipping to production since week two , and the team reviews your PRs the way they review each other's.

Time-to-liquidity is materially down , at the median and at P90, driven by tools you built rather than by headcount.

SLAs are real across ingest, processing, and outtake: published, instrumented by you, tracked, met.

Leadership runs on your dashboard. A weekly Vault operating view people actually open, wired to live station data.

30 to 50% fewer manual handoffs , with exception handling running as a classified queue rather than a chat thread.

The Retool ceiling is gone. The queue and SLA tooling is real software, and the vault team asks you for the next thing rather than working around the last one.

Volume scales without linear headcount. That's the test of whether the system got better or just got busier.

We don't want to hear that you use Claude every day. Everyone does. We want to know what you've built with it. Concretely, in this role:

Put an agent on the exception queue : it reads the intake anomaly, classifies it against the exception taxonomy, pulls the item history, and hands a vault tech a recommended action instead of a mystery

Use document models on cross-border paperwork , customs forms, declared values, carrier labels, insurance schedules, so international outtake stops being a human transcription step

Build the Retool replacement yourself first. Prototype the queue or SLA tool in a working state, put it in front of an ops lead, then spend engineering time on the version that lasts

Stand up an MCP server over vault operations data so "why did P90 move last week, and which station caused it" is a question you ask in a sentence, not a ticket you file

Use agents to turn tribal knowledge into runbooks : watch how the best tech handles a hard intake, and make that a documented, enforceable workflow

Vision models on intake QC , so condition and match discrepancies surface at receiving rather than at outtake, when the cost of being wrong is highest

Claude Code or Cursor as how you actually work through a migration , not autocomplete

If you've built agents, written skills or MCP servers, wired up connectors, or shipped internal tooling your team actually depends on, lead with that.

You ship production code. Not "technical." Not "I can read code." You open the pull request, you take the review, you own the consequence when the queue tool breaks at 6am on an auction week. Full-stack enough to move through an operational surface without a handoff: the internal UI, the API, the query, the job.

You've owned an operational system with real SLAs and unit costs and moved the numbers on it. Throughput, cost per unit, rework, capacity utilization, and how the four trade off against each other. Roughly 5+ years of building, but we're reading the work, not the years.

You own a number today and can name it, its baseline, and where you took it, on the first ask.

Strong data fluency. SQL and Python. You pull your own station-level numbers, and you've built the instrumentation behind them because it usually didn't exist.

A track record of turning messy, real-world processes into software. Not a process document. Software you wrote, that operators use, that survived contact with a bad week.

Physical curiosity. You've stood on a floor, watched the work, and changed your mind because of what you saw. This role includes occasional travel to the vault, and the people who are good at it want to go.

You've built real things with AI , not just used it. Agents, skills, MCP servers, connectors, internal tooling people depend on.

Scrappy and high-ownership. You communicate with precision, you move without hand-holding, and you go find the bottleneck rather than waiting for the brief.

Backgrounds that fit: engineers from fulfillment, logistics, or supply-chain platforms; full-stack engineers who moved toward the operational side; forward-deployed and internal-tools engineers; technical founders in warehouse or 3PL software; product people who never stopped shipping code.

We'll be specific so you can prepare. Before any of it: download the app, put a card up, make a bid, or sell. Every round assumes you've used the product, and we expect your point of view to sharpen as you go.

1. Recruiter screen, 30 minutes via Google Meet. The metric you own today, the last thing you shipped yourself, why Alt specifically, and how you're actually using AI. We'll answer your questions and align on comp.

2. Hiring manager with Leore, our founder and CEO, 30 minutes. This role reports to Leore, so he runs the hiring manager round himself. Go deep on an operational system you owned: the SLAs, the cost per unit, what broke, and specifically what you built with your own hands.

3. Team interview, two 45-minute sessions. The engineers and ops leads you'll ship alongside. How you work in a shared codebase, and how you work with the people on the floor.

4. Build walkthrough, 45 to 60 minutes. Bring one thing you built and shipped, a repo, a PR, a live URL, an internal tool, an agent or skill you wrote, and open it on screen. What problem it solved, what's ugly about it, where AI helped and where it got it wrong. Then we work a live Vault bottleneck together. Rough and real beats polished and hypothetical.

5. Executive interview, 30 minutes. The closing conversation. What you'd go after first, what you'd deliberately not do, and whether this is the job you actually want.

6. Reference check(s), 15 minutes with a former manager.