Home-care operations platform

Healthcare services

CarePartnersAtHome

A multi-tenant operating system for non-medical home care agencies: intake, caregiver screening and matching, scheduling, incidents and family communication, with risk-aware AI inside controlled workflows.

Status
Alpha, running the founder's own agency
Audit
Append-only, delete revoked
Pipelines
11-state requests, 14-state applicants
Scenarios
28 end-to-end journeys
Stack
Next.js, TypeScript, PostgreSQL + Drizzle (RLS), Better-Auth, Twilio, Resend, Playwright, Vercel

The problem

A home care agency runs on a dozen interlocking workflows: families asking for help, caregivers applying and being screened, matching, shifts, incidents, payroll and keeping families informed. Most agencies hold that together with spreadsheets, text threads and memory, which is fragile for the business and unsafe for the people in it.

What we built

  • Three audiences on one platform: an operator console, a caregiver app that works on a phone, and a family portal, all multi-tenant from the first migration.
  • Server-side state machines for the care-request and caregiver pipelines. Every transition is validated against an explicit allowed list before anything is written.
  • A consent ledger that gates every outbound SMS and email, versioned message templates, and a communications layer the rest of the application must go through.
  • SMS-first intake and screening, so a family is answered by a real person by text and a caregiver can be interviewed at their own pace.

Where AI fits

  • Intake summaries from an operator's call notes, with out-of-scope items flagged and follow-up questions suggested.
  • Caregiver screening by SMS: the next question is chosen per turn, and a structured summary of strengths, concerns and risk flags is produced at the end for a person to review.
  • Incident classification and summarization, a moderation queue, and writing assistance for families and caregivers.

Engineering discipline

  • Tenant isolation is defense in depth: every business query runs inside an explicit tenant scope, and row-level security on every business table makes a missed filter fail rather than leak.
  • Append-only audit: business tables cannot delete, records are archived by state transition, and operator page views of sensitive records are audited too.
  • A scenario harness of 28 end-to-end journeys, including a full engagement lifecycle with incident and audit reconstruction, acts as the canary for the whole pipeline.

What it is worth

  • The founder's own agency operates on it, and it is designed for other agencies from day one.
  • Families and caregivers interact through channels they already use, with a real person on the other side and a record of every consent.

The two lanes

Kept deterministic

  • Tenant scope and RLS
  • Consent before any message
  • Pipeline state transitions
  • Audit log

Where intelligence helps

  • Intake summaries
  • SMS screening turns
  • Incident triage
  • Moderation and writing help