Member of the Technical Staff - Data Engineer

Why Join Stand https://www.standinsurance.com/careers/why-stand/: At Stand, you’ll help build a new class of global property protection. We use advanced physics and AI to model catastrophic risk at the asset level, then automate underwriting and mitigation before loss occurs. Insurance is simply the current delivery mechanism. The real product is a scalable risk engine, our Stand World Model https://frontier.standinsurance.com. We stay when traditional insurers exit. We model what others approximate. And we build systems that change outcomes, not just prices. Our leadership team https://www.standinsurance.com/vision includes former successful founders and CEOs from Metromile, PolicyGenius, WePay, and HotelTonight, bringing deep experience in building and scaling high-growth companies. Background: The property insurance industry is built to price loss after it happens. It relies on coarse proxies, backward-looking data, and manual processes, then accepts damage as unavoidable. Stand takes a different approach. We simulate how real-world catastrophes affect individual properties, translate that into actionable decisions, and automate the business around it. The result is a platform that can underwrite what others can’t and operate with far less friction. Role Summary We are building the self-driving insurer: a national carrier whose standard operating procedure is explicit, instrumented, and handed to agents one proven step at a time — so the book grows without the org. Every one of those decisions runs on data, and every one of them is only as good as the number underneath it. That's the precondition this role owns. Not a dashboard. The warehouse, the definitions, and the query surfaces that underwriting, pricing, and leadership all resolve to — and that our agents will read from long before they're trusted to write. Concretely, the platform is four things: A warehouse where every dataset moves at its own speed. Today we have a centralized analytics database in Postgres, rebuilt hourly in full. It works, and it's why you start on the second problem instead of the first. You migrate it: dbt for models, a real orchestrator for scheduling, per-dataset cadence and freshness SLAs, snapshots on the entities whose history matters. Bind state on minutes, vendor pulls and reference data daily. A trust framework and diagnostics console. Reconciliation between each system of record and the warehouse — row counts, premium totals, status parity — running as tests on every load and failing the run, not as a monthly spot check. A console where anyone, not just you, can see whether a dataset is fresh, whether it reconciled, and which dashboards are affected when it didn't. Turning a discrepancy someone noticed into a permanent test should take an afternoon. A shared semantic layer. One registry of typed, versioned, permission-aware entities and metrics. A metric means one thing, defined once, in version control. Deciding what a metric should mean isn't your call; making it easy for an actuary or underwriter to codify it once, so the next person doesn't redefine it in a dashboard, is your job. The agentic query surface is built on this layer and can't see past it: an agent answers in natural language, shows the query it ran, and reaches nothing its invoking user couldn't reach directly. Data that's safe to be creative with. An automated production → sanitized pipeline producing a de-identified but faithful copy of the book — consistent synthetic identities, preserved distributions, intact joins, edge cases kept rather than smoothed away. It becomes the default seed for local, dev, and staging, and the default substrate for anyone prototyping against real-shaped data. Raw PII access becomes the exception with an audit trail. You'll partner closely with Actuarial and Underwriting, who ask the questions, and with Applied Science, whose models both consume this data and produce more of it. Working to understand what insights are missing and build…