Governing Agentic AI in Senior Housing & Care
Illustrative framework · 2-page PDF · Design commitments, not demonstrated deployment.
Illustrative framework · 2-page PDF · Design commitments, not demonstrated deployment.
The single inference boundary in front of SeniorCRE's in-product assistants. Gemini 2.5 is the default tier today through the Lovable AI Gateway. Per-assistant Claude/GPT routing, production HIPAA-vetted PHI redaction with BAAs, and platform-wide approval middleware are on the roadmap — and labeled as such.
When systems disagree, the operator governs.
The AI Gateway describes the target inference boundary for SeniorCRE: one place where model selection, PHI handling, audit logging, and approval gating compose. This page separates what is live today from what is design intent, using the same four-tier proof model the rest of the platform docs use.
Tier 1 — Built (today): In-product assistants scoped per module call Gemini 2.5 through the Lovable AI Gateway. Inference happens inside the tenant boundary against the canonical operational record.
Tier 4 — Advanced / AI-Assisted (roadmap): Per-assistant Claude/GPT routing, the production HIPAA-vetted PHI redaction + tokenization layer, the dedicated AI audit log, the platform-wide needsApproval middleware, BYOK, and VPC-resident open-weights are all on the roadmap. They are not shipped and are not claimed as production capabilities.
| Model | Tier | Use | Notes |
|---|---|---|---|
| Gemini 2.5 Flash (default) | Tier 1 | In-product assistants scoped per module — drafting, summarization, structured extraction | Routed today through the Lovable AI Gateway. Long-context, multimodal, price/performance fit for current assistant volume. |
| Gemini 2.5 Pro | Tier 1 | Higher-reasoning calls inside the same assistants when the default tier is insufficient | Same gateway path; selected per assistant configuration, not per request from the client. |
| Claude 3.5 / 3.7 Sonnet | Tier 4 — Roadmap | Per-assistant routing for long-form policy drafting and dense regulatory review | Architecture supports it; per-assistant routing not yet wired in production. |
| GPT-5 / GPT-4o-class | Tier 4 — Roadmap | Per-assistant routing where structured tool-calling outperforms | Same as above — gateway-ready, not yet enabled per assistant. |
| Open-weights inside operator VPC | Tier 4 — Roadmap | Residency-restricted operators who require inference inside their own cloud | Design intent. No operator contract uses this today. |
| Direct third-party calls bypassing the Gateway | Tier 4 — Roadmap | — | Not permitted by design. The production HIPAA-vetted PHI redaction + BAA path is out of current scope and gets scoped separately before it gets a date. |
Today, in-product assistants operate on the canonical operational record inside the tenant boundary, with operator/tenant isolation enforced by PostgreSQL Row-Level Security under a non-owner connection role. The production HIPAA-ready AI path — subprocessor BAAs, redaction/tokenization before any third-party call, and a dedicated AI inference audit log — is explicitly out of current scope and gets scoped separately before it gets a date. No PHI flows to any LLM until those BAAs are in place.
AI-driven reads stream. AI-driven writes propose. The pattern is enforced inside specific clinical and financial assistants today (for example, medication-order proposals and ledger entries route through human approval before they persist). A platform-wide approval middleware that sits in front of every AI write across every assistant is roadmap work, not shipped.
Compliance posture as of October 22, 2026: HIPAA safeguards are in progress; a BAA template does not establish an executed agreement. Encryption is implemented, independent test pending; SOC 2 is PLANNED, with no certification or report claimed; HITRUST is NOT YET AUDITED; ONC health IT certification is NOT PURSUED; no completed independent penetration-test report is claimed. AWS’s infrastructure assurance reports are not SeniorCRE reports. Review the Trust Center for control-specific status, evidence and limitations; no operator-production assurance is implied.
No completed independent penetration-test report claimed. SeniorCRE does not claim SOC 2 certification or its own SOC 2 report; reports are published only when issued. The AI Gateway controls described as roadmap on this page will land inside that scope as they ship.
That is the target. This page tells you which of those four properties are enforced today and which are roadmap, so a buyer or auditor can map our claims to evidence without translation.
Status as of October 22, 2026 (last modified 2026-10-22)
| Claim | Evidence step | What is true today | Proof to inspect |
|---|---|---|---|
| Operator and portfolio workspace foundation built (roles, hierarchy, entity tree). | Built | Provisioning controls have been exercised repeatedly in controlled SeniorCRE conditions, including the operator onboarding wizard. Not yet performed for an operator in production; no standard duration is published. | Control test record; synthetic or de-identified data; no operator PHI. |
| Single-community acceptance boundary. | Designed | A gate sequence derived from the migration and acceptance model. No community has gone live for an operator, so no observed duration exists. | Written deployment plan and acceptance-gate model. No execution record exists. |
| Portfolio-wide rollout acceptance across multi-community scope. | Designed | A wave-cadence model from the deployment plan. Sequencing depends on community count, system count, data condition, source access, and operator authority decisions. Not a completed rollout. | Written deployment plan and acceptance-gate model. No execution record exists. |
| Connectors to PointClickCare®, MatrixCare®, Yardi®, and QuickBooks®. | Designed | Named-vendor connectors are roadmap, not built. Generic sample mapping does not establish an implemented connector or operator workflow. No named connector is live in operator production. | Build-queue entry with scope and dependencies. No implementation exists. |
| Operator-Controlled Operating Record: source authority, reconciliation, field-level lineage. | Designed | The Operator-Controlled Operating Record is designed and not yet implemented in any community. Authority rules, reconciliation, and field-level lineage are design intent; synthetic examples do not establish working governance. | Written deployment plan and acceptance-gate model. No execution record exists. |
| Clinical configuration: SeniorCRE as clinical system of record, or alongside an incumbent eMAR read one direction only. | Designed | Designed, not built for operator use. SeniorCRE's native medication administration and eMAR are roadmap and not built; the read-only configuration alongside an incumbent eMAR depends on the Operator-Controlled Operating Record, which is designed and not yet implemented in any community. One authoritative medication record at all times is a design rule. | Written deployment plan and acceptance-gate model. No execution record exists. |
| Barcode-verified administration with an automated five-rights check at the point of medication pass. | Designed | Not built. Corrected September 7, 2026: earlier pages, operator training guides, and generated answers described this control as running, which was false. Implementation boundary: four of the five medication scan surfaces open a camera preview with no decoder and match only a manually typed NDC; one mobile surface decodes frames through the browser-native BarcodeDetector API where the browser supports it (Chromium/Android; not iOS Safari, not most desktops) and compares the NDC alone. No decoding library is bundled, no surface verifies resident, dose, route, or time, and no scan result blocks an administration. The five rights are verified by the administering clinician, not by SeniorCRE. | Build-queue entry with scope and dependencies. No implementation exists. |
| Live write-back into operator payroll and scheduling systems. | Designed | Specified and in the build queue. Read-side ingestion only today. | Build-queue entry with scope and dependencies. No implementation exists. |
| Implementation effort required from the operator. | Designed | Deployment is staged, not effortless: platform access, source access, authority rules, reconciliation, security review, and any history migration are scoped work with operator-side effort. Any claim of zero implementation would be false. | Written deployment plan and acceptance-gate model. No execution record exists. |
Status and evidence class as of October 22, 2026. SeniorCRE has no operator-production deployment. Public timing is gate-based and operator-specific; no standard go-live duration is published. Maintained and reviewed by John Hauber, Founder, SeniorCRE, LLC. A medication-safety control was described on earlier pages as running when it was not; that correction is published in full at /medication-safety-claim.
| Capability | SeniorCRE | Single-Model Vendor | Direct SDK Calls |
|---|---|---|---|
| Model choice | Per-assistant routing (roadmap; Gemini default today) | Locked to one vendor | Per-feature, ad hoc |
| PHI boundary | Tenant boundary today; production redaction roadmap | Per provider policy | Re-implemented N times |
| Audit log shape | Operational audit today; dedicated AI audit roadmap | Vendor-defined | N schemas |
| Approval gating | Per-assistant today; centralized middleware roadmap | Vendor-defined | Per feature |
| Failover | Single provider today; cross-provider roadmap | Single point of failure | Per-feature retry logic |
| Cost attribution | Gateway-level today; per-assistant/per-tenant roadmap | Single bill | Hard to attribute |
| BYOK | Roadmap (enterprise) | Rare | Possible but inconsistent |
An in-product assistant makes a request. Product code does not name a model directly.
Today: Gemini 2.5 via the Lovable AI Gateway. Roadmap: per-assistant selection across providers based on configuration and health.
Today: inference stays inside the tenant boundary against the Operator-Controlled Operating Record. Roadmap: redaction/tokenization before any third-party call, under executed BAAs.
The model returns the response. Today the response is consumed inside the assistant; the dedicated AI audit log is roadmap.
Today: per-assistant human approval inside specific clinical and financial flows. Roadmap: platform-wide needsApproval middleware on every AI write.
Today: writes land in the operational audit log. Roadmap: a dedicated inference log (inputs, model, version, output, approver) feeding the same compliance scope.
SeniorCRE is the operating, compliance, and asset-management layer for REITs, family offices, and institutional capital allocators in senior housing & care.