Senior Care / Home Health Admin / Care Coordination Implementation

Use case: implementation detail for the local SMB automation target category “senior care / home health admin / care coordination.”

Default architecture: conservative, human-in-the-loop admin automation. Focus on intake summaries, scheduling exceptions, family-update drafts, caregiver-note summaries, and onboarding/compliance checklists. Avoid clinical decision-making, diagnosis, medication advice, emergency triage, or unsupervised sensitive communications.

V1 principle: treat this as regulated-adjacent workflow support. The system drafts, validates, routes, and reminds; a human approves anything sensitive.

Recommended Tech Stack

Layer	Recommendation	Reason
Language	Python	Best for document extraction, validation, scheduled data processing, and LLM structured-output workflows. TypeScript is still useful for the review UI.
Workflow engine	n8n for consulting pilots; Temporal or Inngest when productizing	Care coordination has reminders, recurring checks, exception queues, and approval gates. Durable retries and audit trails matter.
LLM gateway	LiteLLM	Keeps provider choice configurable and supports per-task routing for summarization, extraction, and drafting.
Agent loops	Pydantic AI or LangGraph only for bounded exception handling	Most workflows should be deterministic pipelines with LLM steps and human review, not autonomous agents contacting families or caregivers.
Document parsing	Azure Document Intelligence, AWS Textract, Reducto, LlamaParse, or Unstructured	Useful for intake packets, referral forms, care notes, PDFs, insurance/admin documents, HR onboarding forms, and scanned/faxed paperwork.
Structured outputs	Pydantic / JSON Schema	Required for intake fields, task queues, caregiver-note summaries, onboarding checklists, and compliance reminders. Free text alone is too risky.
State / files	Postgres with row-level security; Supabase or RDS; S3 or Cloudflare R2 for raw documents	Keep tenants, clients, caregivers, shifts, approvals, extracted fields, and audit history in a durable source of truth. Use privacy-preserving storage for sensitive documents.
Search / knowledge	Postgres + pgvector	Good enough for staff-facing search over policies, care-plan templates, onboarding documents, agency procedures, and prior non-clinical notes.
Auth / tenancy	Clerk or WorkOS + Postgres row-level security	Use managed auth. WorkOS becomes attractive if agencies need SSO/SAML.
Communication	Twilio, SendGrid, Microsoft Graph, Gmail API	Draft and route reminders, family updates, onboarding requests, and shift messages. Keep human approval for sensitive outbound content.
Calendar / scheduling	Outlook Calendar via Microsoft Graph, Google Calendar API, or agency-management-system schedules	Useful for shift exceptions, missed-clock-in alerts, appointment reminders, and caregiver/client availability coordination.
Observability	Langfuse self-hosted where needed + Axiom or Better Stack	Trace prompts, extractions, approvals, and exceptions. For sensitive data, prefer self-hosting or vendors with appropriate privacy/compliance posture.
Evals	Promptfoo or Inspect AI	Test extraction accuracy, summary fidelity, checklist completeness, and whether drafts stay inside policy boundaries.
UI	Next.js review/approval queue; n8n forms for prototypes	Every draft family update, intake summary, compliance reminder, or scheduling exception needs a simple staff review surface.
Hosting	Render, Railway, or Fly.io for non-PHI pilots; AWS for Healthcare, Azure HIPAA offerings, or Google Cloud HIPAA-aligned infrastructure if a client requires formal healthcare compliance	Do not casually put PHI into a hobby stack. If PHI is involved, confirm BAA/compliance requirements before implementation.

Integration Moat

Agency management / home care platforms: WellSky Personal Care, AlayaCare, AxisCare, Homecare Homebase.
EVV / Medicaid operations: Sandata, HHAeXchange. Treat these as integration/discovery targets, not guaranteed APIs.
HR / payroll / onboarding: Gusto, BambooHR, DocuSign eSignature.
Files/forms: SharePoint, Google Drive, Dropbox, Box.
Accounting/billing: QuickBooks Online, Xero, or Codat.
CRM / intake: HubSpot, Salesforce, website forms, email inboxes, and referral portals.

Per-Use-Case Implementation Mapping

Use case	Pipeline
Referral / intake summary	Email/fax/upload → document parser → structured intake schema → missing-information checklist → staff review → CRM/agency-system record creation.
Shift scheduling exception alerts	Schedule pull → clock-in/availability exception detection → LLM-generated staff note with context → coordinator review → Twilio/email outreach draft.
Family update drafts	Caregiver notes + approved care-plan fields → safe-summary draft → policy/sensitivity checks → coordinator approval → email/SMS send.
Caregiver notes summarization	Mobile notes/forms → structured extraction → flag missing/ambiguous items → non-clinical summary → supervisor review queue.
Onboarding / compliance checklist	Applicant/caregiver docs → OCR/extraction → checklist validation → expiration reminders → DocuSign/HR system follow-up.
Policy / procedure assistant	Upload policies, training docs, and templates → chunk/embed in pgvector → staff-facing Q&A with citations → no external/customer-facing answers without review.

Guardrails / Not in V1

No diagnosis, clinical decision-making, medication advice, or emergency triage.
No unsupervised outbound messages to families, clients, caregivers, physicians, or payers.
No PHI in tools/vendors unless the client’s compliance requirements and BAA posture are explicitly handled.
No broad “care AI platform” pitch. Start with one admin workflow and one approval queue.
No custom auth, custom EHR, custom scheduling system, Kubernetes, or multi-region architecture in V1.

Created: 2026-05-10. Implementation drilldown for senior care / home health admin / care coordination. Confidence: medium; validate compliance requirements before implementation.