Go to the plant
Every engineer visits production lines. You cannot build for a factory you have never stood in.
Hiring for cohort-one deployments
Software thatmoves steel,not pixels.
Most engineers never see their code change a physical object. Here, a model you ship on Tuesday adjusts a foaming shot on Thursday and shows up as fewer warranty claims two years later.
What is genuinely hard here
Takt time does not negotiate. If a decision takes 140 milliseconds and the station allows 100, the decision does not exist.
The factory network will drop. The lighting will change between shifts. A new product variant will arrive with no labelled data. A press will be replaced with a different model and the same agent must keep working.
And once the system can write to equipment, every bug is a physical event. That constraint makes the engineering better.
Problems on the board right now
- Sub-100ms multi-camera fusion on edge GPUs
- Few-shot defect learning for new product variants
- Foam density inference from indirect process signals
- Micro-leak signatures below tester detection thresholds
- Sim-to-real transfer for robotic panel handling
- Provable authority envelopes for equipment writes
Open positions
All roles are hands-on. There is no layer of people here who only talk about the work.
Engineering
Edge Runtime Engineer (Rust / C++)
Own the on-premise runtime: deterministic scheduling, GPU inference serving, connector transports and fail-safe behaviour when the network disappears.
EngineeringPerception Engineer (Vision / Sensor Fusion)
Build the models that see enamel defects, foam distribution, connector seating and leak signatures — and know when they are uncertain.
EngineeringRobotics & Simulation Engineer
Panel handling, sim-to-real transfer and digital-twin fidelity good enough to gate a production change.
EngineeringAgent & Reasoning Engineer
Orchestration, evidence chains, confidence calibration and the evaluation harness that decides what is allowed to ship.
FieldForward-Deployed Process Engineer
Live inside customer plants. Own the written success metric, the baseline and the pilot from shadow to advisory.
PlatformPlatform / Infrastructure Engineer
Cloud control plane, signed edge OTA, multi-tenant isolation and the release path from CI to canary line.
How we work
Six rules that decide arguments.
Measure before you claim
Baseline first, always. Unmeasured improvement is a story, not a result.
Safety is a design input
Fail-safe behaviour, bounded authority and reversibility are decided at design time, not patched in later.
Ship narrow, then deepen
One line, one workflow, one metric. Breadth without a proven wedge is how platforms die.
Write it down
Decisions, caveats and failure modes get documented. Institutional memory is a competitive advantage.
Respect the operator
The person on the line knows things the model does not. Their correction is a signal, not an inconvenience.
The system you would be working on
Seven stage agents, a factory orchestrator, a digital twin and an assurance layer — running at the edge, inside real plants, under real takt time.
Form & stampWeld & seamEnamel & paintFoam & insulateWire & assembleLeak & function testRate & pack
How hiring works
Four steps, roughly two weeks, no take-home that takes your weekend.
Intro call
Thirty minutes with an engineer about what you have built and what you want to build next.
Technical deep dive
A real problem from our backlog, discussed the way we would discuss it internally.
System design
Design something under physical constraints — latency, safety, offline operation, reversibility.
Team and values
Meet the people you would work with, and ask us the hard questions about a pre-launch company.
What we offer
Stated honestly for a pre-launch company.
Equity
Meaningful equity — early team, real ownership
Ways of working
Remote-friendly with regular on-site plant time; the plant visits are not optional
Hardware
Edge GPU development hardware and access to real line data from design partners
Learning
Conference budget, and a strong bias towards publishing the engineering we can publish
Stage risk
We are pre-launch. Revenue is not yet proven and the roadmap will change. We will not pretend otherwise in an interview
What partners say about the work
“Our end-of-line functional testers told us what failed. They never told us which press stroke or foam shot caused it. That link is the whole product.”
“Foam voids are invisible until the energy rating comes back wrong. Predicting density distribution before the cabinet cures is the part I could not buy anywhere else.”
“If it writes to the press, it needs an audit trail a quality auditor accepts. Duromex started there instead of bolting it on.”
What the job actually involves
So there are no surprises.
~2
plant visits per quarter, typical
<100 ms
the budget your code lives inside
3
autonomy levels you must design for
0
features that ship without an eval
You will probably dislike this if…
…you want to work exclusively on models and never think about a PLC. …you prefer a clean abstraction boundary between your code and the physical world. …you want a mature product with a stable roadmap.
You will probably love it if you have ever stood on a factory floor and thought: this could all be so much better.
You will fit if
- You are comfortable being measured on a plant's metric
- You think latency budgets are interesting, not annoying
- You would rather ship one proven thing than five demos
- You are happy to be wrong in front of a process engineer
Before you apply
Do I need appliance-manufacturing experience?
For engineering roles, no — curiosity about the physical process matters more. For the forward-deployed process engineer role, yes, appliance or automotive plant experience is essential.
Is this remote?
Remote-friendly, but not remote-only. Regular plant time is part of every engineering role.
What if there is no role listed that fits me?
Write to us anyway with what you would build here. A specific, opinionated message beats a generic application every time.
Come build physical AI that ships.
Tell us what you would build on an appliance line and why.
Duromex is pre-launch. Figures shown are design-partner targets and modelled economics, not audited results. Ask us for the methodology.