01
Context first
Start with users, workflows, risks and operating conditions before choosing the technology.
AI SYSTEMS · SAFETY · GOVERNANCE
Intelligence you can account for.
We build AI systems that can be tested, explained and governed in the real world.
See how it worksApproach
02
Human authority
Keep important decisions with people, backed by clear review, override and stop controls.
03
Evidence before scale
Measure quality, safety, latency and cost on real work before expanding use.
Selected
Projects.
Projects exploring how AI can be used, governed and applied responsibly in practice.
ID
Project
Capability
Year
01
Somali Dementia Awareness
Open recordGenerative AI Video
2026
open project record02
Housing Intelligence Platform
AI Decision Support
2025
Governance
Before it ships. While it runs. When it has to stop.
Every system is governed through three phases, from defining its boundaries before deployment to monitoring it in use and withdrawing it when necessary.
01 · Before it ships
Scope and exclusions
What the system is for, who may use it, what data it may touch, and the cases it must not handle.
Risk and unequal impact
Safety, privacy, bias, accessibility and language, assessed against the people the system actually affects.
02 · While it runs
Named decision owner
A named person approves the output and carries the decision. The system drafts and explains. It does not sign.
Traceable by default
Sources, versions, model changes, approvals and deviations are recorded as the work happens, not reconstructed later.
03 · When it has to stop
Escalation and intervention
A defined route for raising a concern, investigating it and acting, with thresholds set before the system goes live.
Built to be withdrawn
A system can be paused, rolled back, replaced or retired without destroying the record of what it did while live.
Each of these controls leaves a record. Those records, not this page, are the evidence.
Informed by UK GDPR, the Data Protection Act 2018, the Equality Act 2010, the UK's AI regulatory principles, ISO/IEC 42001 and the NIST AI Risk Management Framework. No certification is claimed.
CONTACT