Senior Full-Stack Engineer, Cortex

Everything physical is made from raw materials: polymers, chemicals, metals and minerals, a $16T economy that software has barely modelled. Atmospheric AI computes what those materials should cost, what carbon they carry and where their supply is exposed, from the physics of how they're made, so manufacturers can decide and negotiate from ground truth.

The problem

A procurement lead brings Cortex a question with money on it: what should this grade cost, which of these tender lines is cheapest on a common basis, what happens to my polymer cost if crude moves. She needs an answer she can take into a negotiation. The interface has to be fluid. The answer is assembled for the question, so no question gets a screen of its own. It streams, it takes seconds, it cites where each number came from, and sometimes a tool returns nothing and she still has to know what to trust. That takes product judgement and engineering craft in the same pair of hands. This is the first seat dedicated to owning it end to end.

What you'll own

• The customer-facing application. Interface, API and schema. You decide what a surface shows and what it leaves out, and you defend the call.
• The surfaces of Cortex. Chats, projects, sessions, the analyses they produce and the artifacts they render as. Agents that run on a schedule or a trigger, created by describing them in a sentence.
• Customer onboarding. How a customer's files, grades and plants get into the system, and how their corrections stick, so the second file resolves without asking them again.
• Legible non-determinism. Reserved layout for streaming output, tool calls visible as they land, provenance on each figure, failure states that name the step that failed.
• The front-end bar. You set the standard for everyone who ships front-end code, in-house or partner.
• Building with agents. Most of your code will be agent-written and all of it read by you. You keep the codebase agent-legible and build the tooling that lets the loop run long.

What we're looking for

• You've built and run a B2B application in production. Users, tenancy, permissions, data they couldn't afford to lose, and your hands on both the interface and the backend. We flex on everything else and not on this.
• You have opinions about what a screen shouldn't show. You can name something you removed and what happened next.
• AI has changed how you write code. You direct agents on production work, read everything they produce, know the bugs they get confidently wrong and have built tooling for yourself.
• You've built over slow, partial or uncertain output. Streaming, long-running jobs, sources that disagree. It doesn't have to have been an LLM product.
• You take an ambiguous question end to end. You say what the outcome will be and how we'll know, then land it.

Nice signals: commodities, procurement or materials knowledge (a plus, never a gate); design ability or close work with designers; early-engineer experience; data-heavy interfaces such as large tables, charts and maps.

Stack: Next.js, TypeScript, MUI, Mapbox; Python, FastAPI, Postgres, Dagster, AWS; LangGraph, MCP, Pydantic AI. A signal, never a requirement.

The take-home: given a customer question and the tool results behind it, design the surface that answers it.