Title
Applied AI androboticsengineering.
We build production AI systems and robotics from the Devon and Cornwall border. Some clients are entering the field for the first time. Others have been in it for years and have run out of road.
- Practice
- AI · Robotics
- Sectors
- Seven
- Range
- Fieldbus → foundation models
- Clients
- Gov · research · industry
- Datum
- 50°49′36″N 004°32′38″W
Free tool · no sign-up
South West Autonomy Atlas
Can you legally fly a drone there? Can you put an uncrewed vessel on that water? Drop a pin anywhere in Devon or Cornwall and get a plain-English verdict for your aircraft or vessel class — flight restriction zones, prison no-fly areas, military danger areas, harbour limits and marine protected areas.
The constraint is no longer capability.
For a wide class of bounded tasks, foundation models are good enough. Robotics hardware has collapsed in price. Everything around them has not kept up, and that is where programmes actually die.
Detail A
Integration
Nobody owns the connection into the systems of record. The registry, the property management system, the dealer system, the ledger. It is unglamorous work and it is where most programmes quietly stop.
Detail B
Assurance
Sooner or later someone asks you to prove the thing behaves as claimed. An auditor, a regulator, an insurer, occasionally a court. A system that cannot answer is a liability, however well it performs.
Detail C
Regulation
Sometimes the product you want is illegal where you want to sell it. Better to find that out in week three than after a factory has been paid.
Client register.
Companies whose AI and robotics we built. Every one is in production. Most have a published whitepaper, so you can read the engineering rather than take our word for any of this.
FarmHQOffline-first compliance platform, statutory registry filing, and the aerial robotics programmefarmhq.ai ↗QR-02
QuanMed AIDomain language model, clinical decision support architecture and patient-controlled data custodyquanmed.ai ↗QR-03
XFoneReal-time voice stack under 800 ms median, grounded responses, dealer management integrationxfone.ai ↗QR-04
Sleepless TradesmanReasoning layer, computer vision for site photography, and persistent memory architecturesleeplesstradesman.com ↗QR-05
GuestAIRetrieval-grounded concierge, multi-tenant isolation at the database, PMS adapter layerguestai.co.uk ↗QR-06
BlockAIAutomated market making across 120+ exchanges, cross-chain settlement, on-chain escrowblockmm.ai ↗QR-07
Robot AutoTraderRobotics market analysis and the marketplace platform behind itrobotautotrader.com ↗
ChainAIPlatform engineering and marketing automationFinance · Marketing
Sheet 03/08 — Case annotation · aerial programme
The most valuable thing we produced on one robotics programme was the finding that the client’s intended product could not lawfully be sold.
A client came to us wanting a crop-spraying drone. Our regulatory research established that applying pesticide from the air in the United Kingdom requires a permit from the Health and Safety Executive, and that only trial permits have ever been issued — the obvious product was being sold into a market that did not legally exist. A spray platform also exceeds 25 kg, placing the operator in the Civil Aviation Authority’s Specific Category.
They did not buy a spray drone. They built a software product for aircraft their customers already owned, and reached revenue without a factory. We would rather deliver that finding in week three than a working prototype in month eighteen.
Capability schedule.
Things break at the seams between disciplines, not inside them. Easier to engineer a seam away than to coordinate three suppliers across it — which is why the range below is deliberate rather than opportunistic.
Applied AI systems
Retrieval-grounded assistants, document and vision pipelines, real-time voice agents, agentic workflows with constrained tool use, and the evaluation harnesses to measure all of it.
Robotics & autonomy
Aerial systems and mission software, CANopen fieldbus and motion control, perception and sensor fusion, teleoperation, and the regulatory analysis that determines what may lawfully be operated.
Platform engineering
Multi-tenant SaaS, offline-first mobile, web platforms and APIs — the substrate AI features have to live on if they are to be used by more than one person.
Data & compliance
Integration into systems of record, data residency, tenant isolation, audit logging, and the evidence trail a regulator will ask for.
Research partnership
Study design, ethics and IRB documentation, dataset construction, model evaluation against clinical endpoints, and honest reporting of what results do and do not support.
Brand & communication
Identity systems, design languages and the public artefacts — whitepapers, guidelines, product sites — that determine whether serious work is taken seriously.
Seven industries. The techniques transfer.
What we built for a field outside Bude runs unchanged in a plant room in Dubai. The confirmation loop that keeps a statutory livestock filing safe turns out to be the same one that keeps a clinical suggestion safe.
Where a wrong answer is not a bug report
Diagnostic and predictive modelling from clinical imaging, research-grade data architecture with patient-controlled custody, clinical decision support positioned as augmentative, and the ethics documentation that lets a study proceed. We work to research standards for dataset provenance, reproducible pipelines and reporting that separates what the data supports from what the authors hope.
Built to survive scrutiny.
For government, healthcare and research clients this is frequently what decides the engagement, so we state our position plainly.
The determinism boundary
Every system has a documented line between what is probabilistic and what is not. Compliance rules, statutory calculations, withdrawal periods and tax arithmetic are implemented deterministically. A model may draft across that line. It does not compute across it.
The AI drafts, a human confirms
Nothing we build submits to a government registry, files a statutory return, dispatches funds or commits a clinical decision on its own initiative. The model prepares, a named human approves, the approval is logged.
Client data is not used to train models
Not ours, not a vendor's. Where a commercial inference provider is used we contract for zero-retention and no-training terms, and record it in the client's data protection documentation.
Residency enforced at every layer
Where a client requires UK-only or EU-only processing, that constraint propagates to model inference too. We do not make an exception for the AI. Tenant isolation is enforced at the database, not only in application code.
The infrastructure is already here. The software isn't.
Every transatlantic cable that reaches Britain comes ashore at Widemouth Bay, four miles up the coast from this office. That is not a tourist-board fact. It is the start of a list.
The Met Office runs one of the largest supercomputers in the country out of Exeter. Goonhilly has been talking to spacecraft since 1962. Plymouth builds autonomous vessels and berths the Navy. Newquay has a spaceport. Appledore still builds ships. Torbay has been making photonics since before anyone called it a cluster.
Sixty years of deep-technology infrastructure, in two counties most of the industry treats as a holiday destination.
What the peninsula has never had is a software layer to match it. Engineers come out of Camborne School of Mines and Exeter and go to Bristol, or London, or further. The infrastructure stays. The people leave. We built this company on the view that this is the wrong way round.
Latitude
50°49′36″NLongitude
004°32′38″WGrid ref
SS 2065 0634Road
A39 Atlantic HighwayDevon border
3 milesPostcode
EX23 9JYFig. 07.3 — Coastal section, Celtic Sea to the plateau the practice sits on
The practical reason
There is a less romantic argument for being here. The constraints are real. Build for a farm three miles from usable signal, in weather that takes the power out, for someone who will not read a manual and cannot afford to be down for a day — and you end up with software that survives a hospital basement, a plant room and a rural garage without being asked to. Offline-first was not an architectural preference we read about. It was the only thing that worked.
Read before you call.
QR-WP-001
Capability whitepaper
What we build and how — engineering method, technology platform, robotics practice, assurance position, regulatory posture, named case studies and engagement models. Closes with a note identifying which claims rest on forecast rather than fact.
Read →QR-BG-001
Brand guidelines
The identity system: monogram construction geometry, logo usage and misuse, the palette with measured contrast ratios, typography, graphic language and application standards.
Read →Tell us what you are trying to do.
A conversation, then a short written assessment of what we think the real problem is — frequently not the problem as first described. We do not charge for that stage, and we would rather establish early that we are the wrong supplier than discover it at month four.
Contact us →