Everything we sell ends in a working system: these are the nine lines of work that get it there.
The nine are one practice rather than nine departments to shop between. Eight are set out in the order a system passes through them: deciding what is worth building, laying the ground it stands on, building it, putting it in front of the people who have to use it, proving it to the reviewers who have to accept it, and keeping it right once it is live. The ninth is not a stage but a way of working, for the engagements where the problem is still moving and the answer is an engineer inside your organization rather than a scope agreed in advance. No separate team sits behind any of them. The partners and engineers on your engagement work across all nine, so the boundaries on this page describe the work rather than who reports to whom.
We work with leadership teams on AI strategy: use-case discovery, target architecture and an honest read on build versus buy. You leave with a roadmap and a costed build plan, not a deck that needs another firm to execute it.
AI is only as good as the estate beneath it. We build the data pipelines, model serving, cloud architecture and observability that keep systems running, on infrastructure your own team can operate.
This is the core of the firm: agents, retrieval, copilots, decision systems and the software around them. Every build carries evaluation, security review and observability from the start, engineered for production traffic rather than the demo.
A system nobody adopts is a failed system. Training, runbooks and working sessions with your engineers and operators are deliverables in every engagement, so the capability stays in your organization after we leave.
AI your customers talk to, answering in your name.
We build customer-facing AI: service copilots, assisted support and retrieval over your own knowledge. Each system ships with the evaluation suite that keeps its answers accurate as models and content change.
We design AI systems for enterprise risk: threat modeling, access controls and audit trails from the first architecture. The founders built and operate attack-surface management tooling that reports against SOC 2, ISO 27001, HIPAA and PCI-DSS, and that discipline carries into every engagement.
AI for the finance function, built to survive an audit.
We build decision support, document retrieval and workflow automation for finance teams, engineered for auditability from the start. The founders designed payments infrastructure where every payment is bound to an auditable spend policy, and that standard shapes our financial work.
Keeping the system right after launch, as the models move.
Launch is not the finish of an AI system. Models are revised and retired, data and content move, and the business changes what a correct answer looks like. We keep the evaluation suite, the prompts, the retrieval and the infrastructure current, watch what production is actually doing, and review the roadmap with you rather than waiting for something to break. Where a system needs support rather than only maintenance, we run that too, with AI agents on the volume and our engineers behind them on the escalations.
Senior engineers inside your organization, still accountable to us.
Some work cannot be written into a fixed scope, because the problem is still moving while you describe it. We place senior engineers inside your teams to build it with you, under our review and our security standards rather than handed across to yours.
Strategy decides what is worth building and what should be stopped before anyone writes code. Data and Platform is the ground the rest stands on: pipelines, model serving, cloud architecture, observability. Engineering is the build itself, and it is the core of the firm. Adoption and Enablement faces the people who work for you; Customer Experience faces the people who buy from you, and between them they decide whether the result is used at all. Cybersecurity and Finance Operations read less like products than like standards the build has to answer to, one for your security reviewers and one for your auditors. Managed Operations is what happens after launch, when the models underneath move and somebody has to own the consequence. Leave any of them out and a system still ships. It just does not last.
What a single engagement draws on
A real engagement is rarely one of these. Retrieval over regulated documents arrives as Engineering work, then becomes a Data and Platform question about where those documents live and who is entitled to see them, a Cybersecurity question your reviewers will ask before it goes live, an Adoption question about the people who have to trust what it returns, and, once it is running, a Managed Operations question. You are not handed between practices while that happens. One founding partner is accountable for the whole engagement, from the first conversation through to handover, and the people who scope the work are the people who build it. There is no account layer between you and the engineers writing the system.
What we do not sell
Three things we are asked for and decline. We do not sell people by the seat. No engineer is placed under your management to work a backlog, because the unit we are prepared to be accountable for is a working system rather than a supply of people. We do not resell software licenses. The technologies named on this site are ones we build with rather than vendors we represent, so when the honest answer is buy rather than build, that recommendation does not turn on which name you choose. And we do not sell strategy that ends as a document for another firm to execute. If we would not be prepared to build it, we do not write the plan for it.
Talk to a partner
If you are not sure which of these your problem belongs to, describe the problem rather than the category, and a founding partner will tell you which lines of work it actually needs.
