Schedule — capability

Fieldbus to foundation models.

We are deliberately not a single-stack shop. The right technology for a CANopen device profile is not the right technology for a multi-tenant hospitality platform, and pretending otherwise produces systems that are elegant in one layer and negligent in another.

The platform, by layer

L01

Interface

Web, mobile and voice surfaces; offline-capable field applications; 3D and scientific visualisation

Next.js · React · TypeScript · React Native + Expo · Three.js · real-time speech

L02

Reasoning

Language, vision and speech models under retrieval grounding and constrained tool use

Frontier models · structured output enforcement · task-specific evaluation suites

L03

Orchestration

Agentic workflows, tool permissioning, confirmation loops, deterministic automation

Owned orchestration layers · event-driven rule engines · queue infrastructure

L04

Data

Systems of record, vector retrieval, tenant isolation, audit trails

PostgreSQL + row-level security · pgvector · MongoDB · append-only audit logs

L05

Integration

Connection into the platforms clients already run

Government registries · Guestline · Mews · Apaleo · OPERA · ServiceM8 · Jobber · Commusoft · automotive DMS

L06

Control

Robotics and device-level control

CANopen stacks and device profiles · embedded C/C++ · mission planning · telemetry

L07

Infrastructure

Deployment, isolation, observability

Docker · AWS ECS · Fly.io · Vercel · Turborepo · CI with visual regression

L08

Assurance

Proving the system does what is claimed

Playwright screenshot verification · model evaluation harnesses · smoke suites

Engineering method

Eight rules, each of which has cost us money.

These are not aspirations. Each is applied on live engagements, and each has involved telling a client something they did not want to hear.

01

Regulation before hardware

We establish what is lawful in the target jurisdiction before anything is designed, let alone bought.

02

Own the software layer

Rebranding a third-party platform buys a logo and the manufacturer's liability profile. It does not buy the ability to change behaviour or control the data path.

03

The AI drafts, a human confirms

Nothing we build submits to a registry, files a return, dispatches funds or commits a clinical decision on its own initiative.

04

Build for the network you have

Offline-first: local write, deterministic conflict resolution, sync on reconnection. A system that needs signal to record a fact will not be used when the fact occurs.

05

Ground the model

Assistants answer from retrieved, verified data or they say they do not know. Tool calls are constrained and confirmed.

06

Ship to real users in days

Long release cycles hide wrong assumptions. Ten real users in a fortnight beat a specification written over a quarter.

07

Verify mechanically

Automated tests, model evaluation suites, and browser-driven screenshot verification across layouts. Unverifiable claims eventually stop being true.

08

Say what is uncertain

Every substantial document marks the claims resting on forecast or on a single unverified source.

Engagement models

Feasibility and regulatory assessment · build partnership · embedded engineering · robotics programme · research collaboration. Most long relationships start at the first and progress through the others.

Full detail in the whitepaper Contact