Decisionsupportacitizencouldhaveexplainedtothem
Public bodies carry an obligation most enterprises do not: the reasoning behind a decision affecting a citizen has to be explainable, contestable and recorded.
Theconstraintisthedesigninput
In public administration, the constraints usually treated as obstacles to AI adoption are the same constraints that make a governed architecture necessary. A decision affecting a citizen has to have a legal basis, a recorded reasoning, an identifiable decision-maker and a route to challenge. An AI system that cannot supply those cannot be used for the work regardless of how good its output is.
So the useful starting point is not the model. It is the record: what was considered, which rules applied, what was decided, by whom, and on what evidence. Once that exists, applying AI to the assembly of the case becomes a proportionate and defensible step, because the human decision-maker and the audit trail both remain intact.
Thecaseworkontology
A case, the application it concerns, the policy it is filed under, the caseworker assigned, the objection raised and the evidence produced — the record as a model, not a narrative.
- Caseworker—ASSIGNED_TO→Case
- Casework agent—AUTHORIZED_FOR→Case
- Case—CONCERNS→Application
- Application—FILED_UNDER→Policy
- Review—ASSESSES→Application
- Review—PRODUCES→Evidence
- Objection—CONTESTS→Application
Whereitapplies
Casework support
Assembling the file: applicable rules, precedent, submitted documents and missing information.
Document intelligence
Extraction and structuring of submissions, correspondence and legacy records.
Decision support
Consistent application of criteria, with the reasoning recorded for review and challenge.
Knowledge access
Making policy, guidance and prior decisions findable with citations, for staff rather than the public.
Local models
Inference on models hosted inside the organisation's own boundary, including in national languages.
Audit and transparency
Decision records designed for inspection, including which model and policy version were in force.
Registry documentation
The material an algorithm register or transparency obligation typically requires, produced from the system rather than about it.
Air-gapped operation
Fully disconnected deployment where classification requires it.
Multilingual handling
Working in the languages a public body actually receives, rather than only in English.
A line we hold
The platform is used to prepare and evidence decisions, not to make consequential decisions about individuals without human involvement. Where a decision affects a person's rights or entitlements, a named official decides.
Questions
- Can this run entirely inside our own environment?
- Yes, including fully disconnected with no outbound network dependency, using models hosted locally.
- How do we document this for a transparency obligation?
- The evidence model is designed to produce most of what such obligations ask for: purpose, data used, model and policy version, human involvement and review arrangements. What a specific register requires still needs assessment against your own legal advice.
- Does the platform make decisions about citizens?
- No. It assembles cases and applies criteria consistently so that an official can decide, and it records the reasoning. The decision remains a human act with an identifiable decision-maker.
Continue
- Trust centreResidency, security and procurement material.
- EU AI ActWhere the obligations tend to bite in public administration.
- GDPRData protection controls and lawful-basis considerations.
- Private AI cloudOn-premises and disconnected deployment.
- GovernanceHuman oversight and change control.
- KnowledgeDocument intelligence with citations.
Bring us one operational problem.
You do not need a finished brief. Bring the problem — we will work out the next step together.
Or book a call with the team