Sovereign by design, explainable by obligation
In-country and air-gapped AI for ministries and municipalities: citizen services, case triage, and a decision trail a public body can defend.
Sovereignty
Residency is an architectural property, not a contractual promise: open-weight models inside your jurisdiction, offline mirrors so a disconnected estate can still be patched, and logs showing where every inference ran.
Citizen services
One assistant across web, WhatsApp, SMS, the telephone line and the counter. Voice is what reaches older and rural citizens, so it is built first rather than retrofitted.
Accessibility
WCAG 2.2 AA
tested on every public-facing interface we deliver, including voice and chat
Casework
The recoverable time is in getting the right case to the right desk with the right documents attached. Determinations affecting rights stay with a named officer.
Records
Classification-aware retrieval with page-level citations, and candidate redaction of personal and exempt material for a reviewing officer to confirm.
Defensibility
Every output carries its sources, the rule it relied on and the officer who approved it, in an immutable log that outlives a change of supplier.
Deployment
Models
Regulation
Standards
Channels
3–5 weeks
Which services are automatable, what residency permits, and an indicative architecture you can put straight into a tender.
Feasibility assessment, deployment architecture, procurement-ready requirements
10–16 weeks
One service line built end to end inside the target deployment model, against a measured baseline.
Working service, national-language evaluation set, officer review interface
Ongoing
Rollout across service lines, operation under an SLA inside your environment, and transfer to your own team.
SLA operation, periodic red-teaming, maintained evidence pack, knowledge transfer
Yes, and it is most of what we do. In-country deployment runs open-weight models on government cloud, a national data centre or your own hardware; disconnected sites get offline artefacts and a patch route. We have delivered into the Türkiye Ministry of Health, Turkmenistan eHealth and the Saudi Arabian government.
Every output carries its sources and the rule it relied on, every override is recorded, and the log is immutable and exportable. Determinations affecting rights stay with a named officer. If a use case cannot be made explainable to that standard, we say so before the procurement rather than after the pilot.
When the service is broken rather than slow. If eligibility rules contradict each other, or three departments disagree about who owns an outcome, AI will industrialise the confusion in a plausible tone of voice. Fix the rules and the ownership first. If the policy is a decision table, use a decision table.
Regulator, budget, deadline, legacy core, a board that has been burned once already. Tell us what you are working around.
