“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.”
Answers,by who isasking.
Plant leadership, process engineers, quality, IT/OT and commercial teams all ask different questions. These are grouped that way rather than into one long list.
Operations and plant directors
The questions we hear first from the people accountable for output.
What is the fastest path to a number I can defend?
One line, one workflow, one written metric. Fourteen days of shadow mode gives you an honest baseline, then advisory mode gives you an attributable delta against it.
Will this slow the line down?
It must not, and that is designed in. Station decisions target under 100 milliseconds so the agent operates inside takt rather than alongside it.
What if the system is wrong?
Below its confidence threshold the agent does not act — it holds the unit and escalates with probable cause and evidence attached.
How much of my team's time does a pilot need?
A named metric owner, some process-engineer time during scoping and instrumentation, and IT/OT time for the network and connectors.
Manufacturing and process engineers
The questions from the people who will use it every shift.
How does it handle a new product variant with no data?
Few-shot learning against the appliance knowledge base plus the physics of the process, running in shadow until it has earned confidence on that variant.
Can it really predict foam voids without seeing inside the cabinet?
It infers from shot weight, mix ratio, jig temperature, cure timing and downstream thermal signals. It is a prediction with a confidence, not a measurement, and it is validated against your tester.
What can it actually change on the equipment?
Only what is inside an explicitly granted authority envelope — for example bounded press parameters, enamel setpoints or foam shot parameters, defined with the equipment owner.
What happens to my override?
It is logged as a correction and used as training signal. Engineer corrections are the highest-value data the system receives.
Where each question lands on the line
Most questions turn out to be about one specific stage — which is why the platform is built stage by stage rather than as a single model.
Form & stampWeld & seamEnamel & paintFoam & insulateWire & assembleLeak & function testRate & pack
Quality, test and regulatory teams
The questions that decide whether this survives an audit.
Can we export the evidence for a single serial number?
Yes. Every decision, the evidence behind it, the action taken and the approver are stored in an immutable, queryable, exportable audit log.
How does this help with energy-rating conformance?
By forecasting thermal performance from foam density distribution before the energy test, so non-conforming cabinets are caught and corrected rather than discovered late.
Does it change our non-conformance process?
It feeds it. Non-conformance and disposition records are written back into your quality system with the attribution already attached.
Will an auditor accept an AI decision?
An auditor accepts evidence. That is why every verdict carries citations back to source signals and to the approving human where one was involved.
Infrastructure, security and controls teams
The questions from the people who have to approve it.
Where does the software run?
A K3s-based edge runtime on IGX-class hardware inside the plant, with an AWS control plane for management and learning. The control path has no cloud dependency.
What network access does it need?
Read and scoped-write access to the registered equipment, plus an outbound channel to the control plane. Production continues if that outbound channel is lost.
How is tenancy isolated?
Per-tenant data, model and vector-store scoping enforced by the platform, with role-based access control and full audit logging.
How are updates delivered?
Signed OTA updates, canaried on one line with automatic rollback on regression, gated by CI evaluation and twin validation.
What is the fail-safe behaviour?
Fail-safe stop paths are verified at commissioning. If the runtime is unavailable, the line reverts to its existing behaviour without Duromex writes.
Procurement, finance and executive sponsors
The questions that decide whether it gets funded.
How is Duromex priced?
Per line, per factory, or an enterprise agreement. Details on the pricing page.
What is the typical payback period?
Our modelling targets under twelve months on a single line where the chosen loss is material. That is a model, not an audited result.
What is the risk of buying from a pre-launch company?
Real. Roadmap, hardening and support all mature during cohort one. Design-partner commercial terms reflect that risk explicitly.
What happens if the pilot fails its metric?
We publish the result to you and you stop. A defined exit is part of every pilot scope, which is why the metric is agreed in writing first.
Five things people most often want confirmed
Availability
Pre-launch — design partners and paid pilots
Deployment
Factory edge runtime plus AWS control plane
Latency target
Under 100 ms for station decisions
Data
Stays in your tenant; never trains anyone else's model
Autonomy
Shadow, then advisory, then scoped control — never control on day one
The question people rarely ask out loud
“What if this is just another vision system that detects everything and changes nothing?”
It is the right question. The answer is the trust ladder: shadow mode against a published baseline, advisory mode with attributable deltas, and only then scoped control. If the system cannot beat your baseline, you will know within a fortnight and you will not have granted it any authority.
How you protect yourself
- Agree the metric in writing before instrumentation
- Insist on a measured baseline, not a vendor estimate
- Define the exit condition at the start
- Keep autonomy in shadow until the evidence is there
- Require an exportable evidence chain from day one
What we will not claim
0
audited production results yet
0
customer logos we cannot name
100%
modelled figures labelled as modelled
1
written success metric per pilot
What evaluators say
“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.”
Who is asking these questions
Refrigeration OEM · EU
Laundry OEM · North America
Cooking OEM · Turkey
Dishwashing OEM · Italy
Air-conditioning OEM · Asia
Tier-1 cabinet supplier
Design-partner programme, 2026. Duromex is pre-launch — partner identities are withheld under NDA and shown by segment, not by logo.
Still have a question?
Ask it directly — a process engineer answers, usually within one business day.
Duromex is pre-launch. Figures shown are design-partner targets and modelled economics, not audited results. Ask us for the methodology.