Product Engineer, Infrastructure Deployment

Fluidstack•San Francsisco, CA

5d•$150,000 - $350,000•Onsite

About The Position

Fluidstack is building civilization-scale infrastructure for AI, aiming to deliver 10 to 100s of GWs of compute faster than anyone else. This involves rethinking every layer of the stack, from acquiring power to designing, building, and operating data centers. The company is focused on speed and scale, and is looking for individuals who care deeply about this problem space. The Decision Team is responsible for automating the delivery of gigawatts by turning processes from land to live compute into software, forward-deploying product teams with experts on factory floors and sites, and ensuring every supercomputer is delivered faster than the last by leveraging lessons learned across projects. This role specifically involves building the deployment knowledge graph to capture structured data about site infrastructure, encoding real-world sequencing rules for site bring-up into software to automate scheduling and replanning, transforming the contractor bidding workflow into software based on performance data, generating and verifying per-rack checklists and inspection plans, and automating the closeout process with final test reports and as-builts from the graph.

Requirements

Shipped production code in Go, Python, or TypeScript, and can pick up whatever language the problem requires.
Built with LLM APIs (OpenAI, Anthropic, or open-weight models), created and consumed MCP servers, and shipped with agentic frameworks.
Work with AI coding tools like Claude Code and Cursor every day and get agents producing real work autonomously alongside you.
Spotted a problem no one assigned you, designed the fix, and shipped it to production with minimal direction.
Moved fast without leaving wreckage: systems built under deadline are ones other engineers extended rather than rewrote.
Earned credibility with people who don't live in software, field engineers or cabling crews, and driven adoption of your tool on their turf.
Sweat product and design details, and the tools shipped are ones users chose over their spreadsheets.

Nice To Haves

Structured cabling and fiber (OTDR, optical loss testing).
Rack and cluster bring-up.
IP provisioning.
DCIM tooling.
Time on a deployment site.

Responsibilities

Build the deployment knowledge graph, forward-deployed on live sites beside the Infrastructure Deployment Engineers, so every rack, cable run, fiber link, QA inspection, OTDR trace, optical loss result, and copper certification lands as structured data tied to the exact link it tested, and as-builts become a live, queryable model of what was actually built.
Encode the real sequencing rules of a site bring-up (the network room comes up first, the end-of-row rack is the synchronization point) so software generates the low voltage contractor schedule, milestone tracking, site access, and material staging, replans when a shipment slips, and flags the delay before it costs the date.
Turn the contractor bidding workflow into software: vendor bids in, a selected contract out, and every vendor ranked by measured speed and QA accuracy per region, so the next award is made on performance data the graph already holds.
Generate per-rack checklists and multi-stage inspection plans straight from design documentation, verify submitted results against spec automatically, and auto-create the punch list, so deployment engineers spend their time adjudicating exceptions.
Assemble closeout the moment work completes: final test reports, as-builts, and root cause analyses produced from the graph, so handover to operations is a state transition with evidence attached and an ops tech can localize a down link (transceiver versus cable versus DWDM) with zero training.