Case Study
A contractor that secures OT environments for defense, nuclear, and utilities clients.
Heard it coming
1
For a contractor securing operational technology for defense, nuclear, and utilities clients, an asset failure is not a ticket in a queue. It is an outage in an environment that does not get to have outages. The job is exactly what it sounds like: know when a physical asset is about to fail, before it fails.
The client's digital twin system already watched the estate closely: digital replicas of physical assets, real-time monitoring, preventive maintenance, work orders, telemetry, asset health. It could tell anyone what an asset was doing at any moment. The ask was the harder tense: what the asset is about to do.
2
The most expensive warnings in an OT environment are often the ones already being said out loud. The building management system logs the slow drift: temperature, power draw, cycle counts. But the clearest early signals rarely reach a system at all. They arrive in a phone call between technicians, an email after a shift, a remark on the radio: the gauge that has read a little high all week, the noise a machine started making on Tuesday, the safety mechanism that wore through on the night shift and was swapped without a note. The people closest to the equipment report these things every day. They almost never know they are describing an early signal, and never what it means for the configuration of the asset.
That is the gap in the configuration management cycle. Weather and wear do not file work orders. Wind load, salt air, freeze-thaw cycles, the extra duty one shift puts on a bearing, the lubricant swapped during a shortage: each one shifts an asset away from its recorded configuration, in the field, out loud, and the record never catches up. The machine data sat in one place, the human and communication data in another, and configuration management was reconciling against a snapshot reality had already left behind. Every signal was already on site. No system was reading them together.
3
Origami rebuilt the client's proprietary stack in 7 days and extended it to include real time conversational analytics. Three products did the work.
ACI is capture hardware, installed in the client's environment. It turns the conversations that carry the early signals into structured records: who spoke, what was said, the transcript, the consent record, each one signed and tamper- evident. Phone calls and ambient conversation arrive as data instead of evaporating between shifts. Email enters the same model through SDO's ingest, alongside the telemetry.
Vision Builder did the rebuilding. Its agents take the work from design to shipped code, and one of them tests the result the way a user would before anything counts as done. The rebuild and the extension to real time conversational analytics are machine work, and 7 days is what that looks like.
SDO is where the fusion happens. It ingests the full input set together: BMS telemetry, ambient conversation, phone calls, emails, and other communications, and learns the structure from the data itself. One model, all the signals at once, with the human and communication layers arriving on the same clock as the machine data.
4
We proved it on the client's own history. Streaming historical data back through the rebuilt stack in chronological order, the signals that had been reported and lost surfaced again in one view: the drift in the building management readings, the remark about a noise, the wear a shift put on a safety mechanism, the weather load accumulating on an exterior component. Machine data, ambient conversation, and the phone and email traffic line up on one timeline against one asset, and configuration management can finally see what the field has been saying. A signal counts wherever it originates: a BMS trend line, a phone call, an email, a remark in passing. That is the tense the model answers in: what the asset is about to do, not what it is doing.
5
The gap this closes is not exotic. It is the phone call that never became a work order, the shift pattern nobody folded into the maintenance baseline, the weather that quietly aged a component while the record stood still. In an environment that cannot afford an outage, those are the expensive misses. Reading them together is multi-source fusion across OT data, and it is the kind of engagement that runs for quarters.
The thesis held on the client's own history: the stack was rebuilt in 7 days, extended to real time conversational analytics, and demonstrated end to end by streaming historical data in chronological order. It is proven and not yet deployed: the client was not ready to implement. We demonstrate the capability first, and the deployment follows the client's readiness.
Tell us what you integrate and who you've shipped for. Certification starts when your technical leads talk to ours.