AI you can show to your auditors
We build the controls that make AI safe to run in production - identity-scoped data access, approval steps, audit logs, evaluation suites and monitoring - into the architecture, not into a policy document.
- IdentityEvery query runs as the user
- PermissionsEnforced by the data platform
- GuardrailsLimited tools, approvals
- EvaluationRuns on every change
- Audit logWho, what, which model, cost
AI Governance & Evaluation
What we build
- 01
Identity-scoped access
Per-user service principals instead of shared credentials.
- 02
Evaluation suites
Real tasks, failure cases and regression runs before every change.
- 03
Guardrails for actions
Limited tools, approval steps and handling of untrusted content.
- 04
Audit logging
Who asked what, which data and model were used, and the cost.
- 05
Residency choices
Amazon Bedrock or Google Vertex AI when data must stay put.
- 06
Reviews of live AI systems
Access, evaluation and logging reviewed, with a prioritised fix list.
Results from our case studies
Selected work
Built for clients, running today
“The team is highly skilled and adaptive, very disciplined, and takes feedback well. They deliver on time and interact professionally. We continue to learn a lot from them.”
Dan StephensonDirector of Engineering, Data & Automation, Barton MalowHow we work
From first workflow to production
- 01
Map data, users, actions
What the system can read and do, and for whom.
- 02
Enforce in the platform
Access control in identities and data, not in prompts.
- 03
Measure before & after
An evaluation set that runs on every change.
- 04
Document for oversight
Controls and test results for legal, security and risk.
Access has a boundary.
A different role. A different set of permissions.
| Read access | Support | Finance | People |
|---|---|---|---|
| Customer records | |||
| Invoices | |||
| Personnel files |
Select a check or lock to see the decision
Support reading Customer records
Read permitted. Scoped to this role.
Is it a fit?
Who we build this for
A good fit if you have
- AI touching sensitive or regulated data
- Agents that change records, spend money or contact customers
- Teams that must explain how their AI behaves
Probably not for
- Legal opinions - we work alongside your advisers
- Policy-only governance with no system to build in
Technologies we use
- Per-user service principals
- Evaluation suites
- Audit logging
- Amazon Bedrock
- Google Vertex AI
- Model Context Protocol
FAQ
Questions we often hear
Something else on your mind? Ask the engineers who would build it.
Ask us directlyNext step
Let’s talk about your project
Tell us what you are working on, and the engineers who would build it will get back to you.