The Role:
Functionally the Vault is a high-end warehouse with sophisticated internal software and a marketplace bolted directly to it. This is a builder seat, not a product management seat. You own the Vault from arrival to exit: ingest SLAs, intake and routing, exception handling and QC, outtake, international shipping, customs, insurance, carrier integrations, and the tooling that runs all of it. And you own it the whole way: you stand in the room and watch the bottleneck, you pull the station data yourself, you write the tool that removes it, you ship it, you watch what P90 does, you keep it or you kill it. What this role is not: writing PRDs for someone else to implement, running a quarterly roadmap ritual, managing an ops pod, or translating between the warehouse and engineering. If that's the shape of your last five years, this probably isn't the right fit, and we'd rather say so up front than waste your time. You report directly to the CEO. If you want a CEO in the build with you, this is that. If you want distance from the top, it isn't. The metric you own: Time-to-Liquidity (median and P90). Supporting KPIs: ingest and outtake SLA adherence, cost per item processed (labor minutes, rework, fully loaded unit cost), throughput per FTE and items per week, error and rework rate, inventory activation rate (vaulted to live to sold), ship-and-decide velocity (tools and changes shipped per week), and Vault-related seller satisfaction. What You'll OwnThe end-to-end lifecycle, in code. Receiving and ingest, processing and routing, exception handling and QC, outtake and customer handoff. Define measurable SLAs at every stage, instrument them yourself so the numbers exist, enforce them, and hunt bottlenecks systematically instead of working around them. The station-level data that tells you where P90 went does not exist yet in the shape you'll need. You build that too. The Vault tooling, built not requested. Queue management, prioritization, labeling, SLA tracking, the exception taxonomy and the software that enforces it. Retool got the operation this far and it is now the ceiling. Replace it with durable software, starting with the version you write yourself and put in front of an ops lead this week. You'll be in Python, Postgres, and AWS, with React and TypeScript for the surfaces the vault team actually touches. International shipping and logistics. Customs, insurance, labeling, carrier integrations. Cross-border is where delay and risk concentrate. Make it fast and boring, wire it into the SLA and liquidity picture, and put document models on the paperwork so outtake stops being a human transcription step. Inventory activation and liquidity. Shrink the distance between "in vault" and "live on marketplace." Find where activation stalls by item tier, seller segment, and station, then build the product that closes it. Partner with Pricing, Auctions, CX, and Risk so vaulted inventory gets listed and monetized quickly. Quality, reliability, and trust. Error rates, rework, and the exception taxonomy are yours. A mis-handled high-value item isn't a ticket. It's someone's money and someone's trust, and the control that catches it is software you wrote. The bets and the trade-offs. A short quarterly plan tied to time-to-liquidity, cost per item, and activation rate, and a weekly record of what actually shipped. Work shoulder to shoulder with Ops, Engineering, Pricing, Marketplace, and CX. Support exploration of Vault-as-a-Service without letting it distract from core execution. Communicate the trade-offs to the CEO plainly. What Great Looks Like (6 Months) 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. How You'll Use AI HereWe 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. What You BringYou 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-deploy
