Our Technology
A hospital needs AI that behaves the same tomorrow as it does today, keeps its data where the hospital decides, and leaves a record of every step. This is the technology that does that, explained for the people who run hospitals.
The principle
Language models are powerful, and they change every year. What a hospital depends on must not change with them: who the system is acting for, what it may see, what it remembers, and who approved its work. We keep those in an operating layer, and use models as the work requires.
Six parts, one platform
01 · Operating layer
The operating layer that keeps identity, memory, permissions, evidence and audit continuous while models change. Models think; Kurahi keeps everything around them whole, and keeps working between tasks when no model is running.
The model is not the agent.
02 · Persistent agents
Long-running agents that keep their memory, role, policy and unresolved work across sessions and across model changes. The agent supporting your discharge team on Monday still knows on Thursday which cases were waiting for a result, what was agreed and who approved it.
03 · Memory
Structured memory (facts, decisions, open tasks) and narrative memory (what happened, in order), each with its provenance. For every task the runtime retrieves the minimum authorised context, never the whole archive, and records which memory was used.
A model receives enough for the task, never the whole archive.
04 · Safety spine
Every consequential draft is checked by a second, independent model that looks for unsupported statements, contradictions and anything outside the professional scope. Then the responsible professional reviews, corrects and approves it.
Only approved work becomes a record, with every reviewer, timestamp and change kept.
05 · Model choice
Frequent and sensitive work, such as classification, extraction, routing and Japanese clinical documentation, can run on models inside your own environment. Broader models are used only where your policy allows it.
When a broader model is used, the runtime sends only the minimum authorised context and records that it did so.
06 · Where it runs
Data stays in the country: Japanese data in Japan, Brazilian data in Brazil. The deployment decides where data travels, and we state it plainly in each customer’s agreement.
For your hospital
Models can be changed without changing how your teams work.
Roles, permissions and records survive a change of model.
On your premises or on national public infrastructure.
Which models, which evidence and which person approved.
We will walk your IT and clinical leads through the deployment options, the controls and the record it keeps.