01AI systems
AI systems
Agents, retrieval and language applications that do real work inside a business, not demonstrations.
- AI agents
- Retrieval & knowledge systems
- Messy document processing
Public, versioned, self-assessed
We build AI systems, engineered to a published standard.
Whole systems, end to end. We start with how the work actually happens, and hand over something your own team can run.
The problem
Most organisations cannot show how their systems work.
The work was never really visible. Leadership believes it runs through the platform; in practice a good deal of it runs through spreadsheets, re-keyed documents and people who know which fields to ignore.
AI has been layered on top of that, quickly, and often without controls. Software now makes decisions inside processes nobody had mapped, with permissions nobody set deliberately and a record nobody kept.
That was survivable while nobody was asking. It is not any more. Regulators, customers and auditors now expect organisations to produce an inventory of the AI they run, evidence of how it was built and tested, and a record of what it did. Most cannot.
This is an engineering problem before it is a compliance problem. Controls that were never built cannot be documented after the fact.
What we do
Whole systems, in six areas.
Most projects draw on more than one. A system that only exists in one of them does not run in production.
02Product engineering
Product engineering
The interfaces people actually use — internal tools, review dashboards and the consoles that keep a system supervised.
- Internal tools & operational web apps
- Human-in-the-loop review dashboards
- Operator & oversight consoles
03Automation & integration
Automation & integration
Connecting the systems you already run, so nobody reconciles anything by hand.
- System plumbing & integration
- Workflow automation
- Integration into daily tools
04Data & analytics
Data & analytics
The platforms, pipelines and definitions that reporting and AI both depend on.
- Data platforms & warehousing
- Pipelines
- Data quality & governance
05Cloud, delivery & engineering practice
Cloud, delivery & engineering practice
The foundation everything runs on, and the pipeline that gets it there verifiably.
- Foundations: zero to production
- Delivery pipelines
- MLOps
06AI governance & compliance
AI governance & compliance
The documentation, logging and oversight that let you evidence what your systems do.
- Conformity-ready build & governance setup
- Article 50 transparency build
- Decision records & human review
How we build
We publish the standard we build to.
Most firms describe their engineering practice in adjectives. Ours is written down, versioned, and public — the same document we work to and the same one you can hold us to.
It covers how we handle your data and credentials, how code is reviewed and released, how AI systems are scoped and tested, and what happens after we leave.
Built against the standards the industry actually uses:
OWASP
Top 10 for LLM Applications and for Agentic Applications
NIST SSDF
SP 800-218, and 800-218A for generative AI
SLSA
Build Level 3
NIST AI RMF
govern, map, measure, manage
CIS
cloud benchmarks
What you receive
Every project ships with these as standard, not on request:
- Source and infrastructure as code — in repositories you own
- Architecture documentation — what depends on what, and why
- Threat model — what the system touches, and what happens when it misbehaves
- Dependency inventory — a software bill of materials for every release
- Build provenance — signed records of how each artifact was built, verifiable with your own tooling
- Security review record — what was tested, what was found, what was done
- Runbook — how to operate it, written for your team
You own all of it. None of it is contingent on continuing to work with us.
Industries
Built for sectors that still run on paper.
The same engineering, applied where the manual work is heaviest and the tolerance for error is lowest.
Start with what you already run.
We work inside your existing stack — your cloud, your data platform, your model providers, your tools. The first thing we do is understand how the work runs today.
Your platforms, your accounts, your repositories. Everything we build is handed over and runs without us.