- Home
- Business Units
- AI & Digital Infrastructure
Technology for the
Next Industrial Era.
We are not an AI vendor. We do not sell implementation projects, licences or advisory hours. We are a trading and engineering company that decided to research this technology properly on its own operations — and what follows describes what we have found, not what we would like to sell you.
A research programme, run on live traffic
Artificial intelligence is one of our strategic development priorities, and we treat it as a research programme. Our specialists test different schemes for applying and embedding AI into real business processes: procurement, documentation, engineering support, market analysis and internal decision-making. Some of those experiments have gone into permanent use. Others were abandoned once we measured them honestly, which we regard as the more useful half of the exercise.
The work runs on live commercial traffic in a demanding compliance environment. Our agents parse RFQs carrying hundreds of line items, cross-check shipping documents against purchase orders, monitor markets and prepare compliance files — work that is low in value per item and high in consequence when it goes wrong. Every agent has a human sign-off point on anything that binds the company.
Running the technology against real consequences, rather than in a pilot sandbox, is what tells you where it actually holds.
On the limits
Model behaviour is uneven on edge cases, verification remains the expensive part, and the gain is almost never in replacing a specialist — it is in removing the mechanical work that stands between the specialist and the decision. We continue investing on the basis that digital technology will change industrial operations through workflow rather than through spectacle.
What we are researching and testing
Agent architectures for commercial intake
Quotation intake, specification matching and priced response drafting — the path from an inbound RFQ to a reviewable offer.
Automated document verification
Checking commercial and technical documents against purchase orders and specifications, and flagging the mismatch before it becomes a claim.
Market and lead-time monitoring
Continuous monitoring across the product groups we trade, so a change in availability reaches the desk before it reaches the customer.
AI support for engineering workflows
Specification review, calculation checking and technical search — assistance to an engineer, with the engineer still responsible.
Private and on-premise corporate AI
Closed systems for organisations whose data cannot leave the perimeter. Usually this is the condition that makes adoption possible at all.
Human oversight design
Where sign-off belongs, and how to make it meaningful rather than ceremonial. This is the part most implementations get wrong.
Where it is already in use
Aviation & chemicals desks
First-pass quotation handling and document verification, with a specialist reviewing everything that binds the company.
Procurement
Supplier and counterparty screening files prepared and assembled before human review, not instead of it.
Engineering support
Technical search and document cross-referencing across specifications, drawings and certificates.
The hardware underneath
AI has to run somewhere. We specify, build and operate that layer for our own systems — sizing compute against the workload, levelling firmware, and testing before anything goes into service. It is infrastructure we run, not a catalogue we sell from.
Compute we build on
- Gigabyte
- Dell
- IBM
- Supermicro
- HP
Storage we build on
- QSAN
- IBM
- H3C
- Huawei
Networking we build on
- Juniper
- Huawei
- Broadcom
- Compute sized against the actual workload, including GPU inference and training
- Staging, firmware levelling and burn-in before anything carries live traffic
- Optical transceiver and cabling compatibility checked by our own engineers
- On-premise deployment where data is not allowed to leave the perimeter
Curious what an agent could take off your desk?
Describe the process that eats the most hours in your week. We will tell you whether it is a good candidate — and if it is not, we will say so.