Lead Platform Engineer - On-Site (Irvine)

Eggs Unlimited•Irvine, CA

9d•$180,000 - $210,000•Onsite

About The Position

We leverage AI to bring food to every table. You are a hands-on platform engineer embedded with operations. You define the platform the business runs on and build every layer of it: the product a desk uses, the infrastructure under it, and the harness the team ships through. You own the outcome, not the component. AI-assisted development is how the work gets done. Human judgment, controls, and production quality are not negotiable. We are building the intelligent operating platform behind a global food supply chain, combining leading models, our own operational data and business rules, and the judgment of the people who trade for a living. Extracting an order from an email is no longer the hard part. The loop is. Forecast where the market is going, project demand forward by customer and by product, and know what we are long and short on any given day. Then decide what to buy and what to sell, build the load, and book the freight against a computed view of what that lane is actually worth. Trading, supply, freight, and the ledger stop being separate systems stitched together by a spreadsheet and become one connected flow, with a named person accountable at every point where money moves. The product layer is largely greenfield. You will be building it, not maintaining someone else's version of it, though it has to work with the operational systems the business already runs on. We move quickly and we build practical solutions that scale, not demos. None of that loop works in pieces: a forecast is only as good as the data under it, and a booked load has to reach the ledger intact. Somebody has to design the platform that makes it one flow, and build it. That is you. The Orchestrator The person in this seat does not draw a diagram and wait. They take an idea, direct a fleet of agents at it, and produce the working system themselves, at a volume of output that would not have been possible from one person two years ago. If your monthly token consumption would look alarming to a traditional engineering manager, you are probably working the way we work. What we measure is not the volume: it is whether you decompose work well, keep context under control, catch the plausible and wrong answer, and stand behind what ships. You do not need prior experience in our exact stack. What matters is whether you can build things, whether you take ownership, and whether you can hold the big picture while you do it. Think like a customer, act like an owner. This seat pairs that with a solution architecture background. You sit with the traders, the purchasing coordinators, and the logistics desks to see how the work actually crosses systems, then build every layer of the answer yourself. The architect builds here. We are not looking for someone who produces a design and hands it to somebody else to implement.

Requirements

Demonstrated experience designing and building systems that span more than one application: data models, integration contracts, infrastructure, and the failure behavior between them, in real operating environments.
You ship production code. You are accountable for what it does in production, however it was produced. Product features, infrastructure, and harnesses are all real software, and all of it is held to that standard.
Serious agentic development practice. Show us real parallel-agent work: how you decompose tasks, manage context, review and reconcile output, design tests and evaluations, an example of a failure you caught, and a case where you decided not to use an agent. Anonymize anything confidential.
Ownership. You have carried something end to end and been accountable for whether it worked, not for whether it shipped.
Enough backend, data, cloud, and deployment competence to stand up services and the environments they run in, integrate systems, and debug production independently.
Judgment about correctness in nondeterministic systems: evaluations, observability, failure-mode thinking, guardrails, human escalation, and rollback.
You can make technical decisions and tradeoffs understandable to people who are not engineers, and write them down so they outlast the conversation.
Experience building within messy operational reality, email, spreadsheets, chat threads, and phone calls, rather than replacing it with something tidier that operators will not adopt.
Seniority here is calibrated by scope, judgment, autonomy, and outcomes. Show us what you have built.

Nice To Haves

Food distribution, logistics, retail and CPG, trading, ERP or EDI, financial operations, or other trading-partner ecosystems.
Ambiguous, high-ownership environments where designing, building, and running the system were the same job: founder, early employee, solution architect who stayed hands-on, forward-deployed engineer.

Responsibilities

Decide what the platform has to become three capabilities from now, and build today's piece so nobody has to tear it out. Think big, and ship the first slice this month.
Take a validated need to monitored production yourself: the data model, the services, the integrations, and the screen or channel an operator actually works in.
Environments, pipelines, data stores, event flows, identity, and the contracts with the ERP core, partners, and customers, built as code. You decide what the platform is and build it; the engineering lead decides what it has to meet, and reviews it against that.
The substrate the team and its agents ship on: fixtures, replay of real operating history, test data, and the CI scaffolding that lets everyone move fast without shipping blind.
Decide what happens at every seam before it happens: retries, fallbacks, reconciliation, and the handback to a person. Money moves through these seams, so the numbers have to tie out to the ledger.
When something goes wrong where a desk meets it, fix it, whichever service, system, or vendor it passed through, and then close the gap that let it happen so it does not come back. "That is the other system" is not an answer here.
Decompose a problem into work that can run in parallel, direct agents at it, control context, reconcile conflicting output, test what comes back, and remain responsible for the final system. This is an operating capability, not a tool preference.
Ship maintainable, tested, observable, secure systems, with explicit evaluations, audit trails, rollback paths, and safe failure modes. Agent-generated work is held to the same production standard as anything you type.