01 2 to 3 weeks
Assess
We map the system, its data, and its risks, then agree what a production pilot must prove.
Deliverables
- Architecture risk register
- Prioritized delivery plan
Service line
Enterprise AI engineering that survives your security review.
Our enterprise AI engineering practice designs, builds, and operates AI systems that run inside regulated and safety critical operations.
Problems we are called in for
Where AI programs get stuck.
-
A pilot works, and the security review will not sign it off
The model performs in a notebook, but nobody can show how data moves, who can reach the endpoint, or how prompts are filtered. We produce the evidence the review asks for.
-
A model is live, and nobody owns it
The team that built it has moved on. There is no evaluation baseline, no drift check, and no runbook. We put operational ownership in place.
-
An AI feature has to run next to safety critical systems
The system must fail safe and leave an audit trail. We design the boundaries, the fallbacks, and the human checkpoints before the model goes near production.
-
Leadership wants an AI program, and the risk team wants a framework
We map the program to NIST AI RMF and ISO/IEC 42001 as working references, so governance and delivery use the same plan.
What we deliver
Documents your reviewers and operators can use.
- Architecture risk register
- Every component, data flow, and trust boundary, with the risk each one carries.
- Threat model
- Attack paths against the model, its data, and its integrations, with mitigations.
- Model evaluation report
- Accuracy, robustness, and failure modes measured against agreed thresholds.
- Runbook and handover
- How to operate, monitor, roll back, and retrain the system, written for your team.
How we work
Four phases, each with named deliverables.
Each phase has a fixed scope and ends with named deliverables your team keeps. Durations are defaults, agreed for each engagement.
-
- 02 6 to 10 weeks
Pilot
We build the smallest version that runs on real data under real controls.
Deliverables
- Working pilot in your environment
- Threat model
- Model evaluation report
- 03 4 to 8 weeks
Harden
We close the findings from your security review and prepare the system for production load.
Deliverables
- Closed review findings
- Monitoring and alerting
- Rollback procedure
- 04 Ongoing
Operate
We run the system with your team, or hand it over with everything needed to run it.
Deliverables
- Runbook and handover
- Service reviews
Standards and tooling
The references we work to.
Naming a standard means we use it as a working reference. Our certifications are listed on the Trust Center .
| Reference | What it covers | How we use it |
|---|---|---|
| NIST AI RMF 1.0 | AI risk management | Structures the risk register and governance plan |
| ISO/IEC 42001 | AI management systems | Reference for roles, controls, and review cycles |
| OWASP Top 10 for LLM Applications | Security risks in language model systems | Checklist for the threat model and test plan |
| MITRE ATLAS | Adversary techniques against AI systems | Source of attack paths |
Procurement questions
What your team will ask before the first call.
What happens in the first phase?
The first phase is Assess. It runs 2 to 3 weeks by default and produces an architecture risk register and a prioritized plan. You decide whether to continue after it.
How does your work get through our security review?
The threat model and architecture risk register are written for your reviewers from the start. The Harden phase closes the findings they raise before anything reaches production.
Which standards do you work to?
NIST AI RMF, ISO/IEC 42001, OWASP Top 10 for LLM Applications, and MITRE ATLAS, used as working references. Naming a standard is not a claim of certification against it.
Where can we see your own certifications?
The Trust Center lists every certification and audit report we hold, with its scope, auditor, and dates. Anything not listed there is not claimed.
What do we receive at exit or handover?
A runbook and handover package covering operation, monitoring, rollback, and retraining, so your team can run the system without us.
Bring us the program your team is worried about.
A senior engineer replies within one business day. The first call is a technical conversation, not a sales presentation.