Medical / Dental / Veterinary / Therapy Practices Implementation

Use case: implementation detail for the local SMB automation target category “Medical, dental, veterinary, and therapy practices.”

Automation focus: Front-office automation for reminders, intake forms, referral/document routing, post-visit instruction drafts, procedure assistants, and no-show recovery — without clinical decision-making.

V1 principle: automate intake, extraction, routing, reminders, and draft generation first. Keep humans in the approval loop for customer-facing, financial, legal, clinical, contractual, or compliance-sensitive outputs.

Common Stack Pattern

Layer	Recommendation	Reason
Language	Python for automation/extraction; TypeScript for UI where useful	Python is strongest for data/document/LLM workflows; TypeScript is strong for web review surfaces.
Workflow engine	n8n during consulting; Temporal or Inngest when productizing	n8n is fast for SMB connectors and pilots. Temporal/Inngest add durable retries, schedules, human approvals, and audit trails.
LLM gateway	LiteLLM	Provider swapping and per-task routing without rewriting code.
Agent loops	Pydantic AI or LangGraph only where needed	Most SMB automations should be deterministic pipelines with LLM steps, not autonomous agents.
Structured outputs	Pydantic / JSON Schema	Use schemas for extracted fields, checklist outputs, and draft metadata; never depend on free text alone.
State/files	Postgres via Supabase or RDS; S3 or Cloudflare R2 for documents	Single source of truth plus durable raw document storage.
Search/RAG	pgvector in Postgres	Good enough for SMB-scale citations over policies, templates, customer docs, manuals, proposals, or service histories.
Auth	Clerk or WorkOS	Do not build auth. Use WorkOS when SSO/SAML is needed.
Observability	Langfuse plus Axiom or Better Stack	Trace LLM calls, extraction failures, review outcomes, and workflow exceptions.
Evals	Promptfoo or Inspect AI	Test extraction accuracy, draft quality, policy boundaries, and edge cases before relying on automations.
Review UI	Next.js; n8n forms for early pilots	Most automations need a lightweight human approval queue.
Hosting	Render, Railway, or Fly.io	Enough for SMB pilots; move to AWS, GCP, or Azure only when required.

Integration Moat

Practice systems: athenahealth Developer Portal, Epic on FHIR, Oracle Health/Cerner Developers, Open Dental API, Provet Cloud API. Availability varies by practice and vendor.
Healthcare interoperability: HL7 FHIR, HL7 v2.
Forms/docs/comms: Google Forms API, Typeform API, Twilio, SendGrid, DocuSign API.
Hosting/compliance references: AWS for Healthcare, Azure HIPAA offerings, Google Cloud HIPAA-aligned infrastructure.

Per-Use-Case Implementation Mapping

Use case	Pipeline
Appointment reminders + no-show recovery	Schedule pull → reminder rules → message draft/template → send via approved channel → no-show follow-up task.
Intake form summarization	Patient/client forms → structured extraction → missing fields → staff review → practice-system update where allowed.
Referral/document routing	Inbound fax/email/PDF → classify referral/lab/insurance/admin → route to queue → staff checklist.
Post-visit instruction draft	Clinician-approved template + visit metadata → draft communication → clinician/staff approval → send.
Internal policy assistant	Policies/templates/procedures → cited staff Q&A → no patient-specific diagnosis or treatment advice.

Guardrails / Not in V1

No diagnosis, treatment recommendation, medication advice, or clinical triage.
No PHI in tools without confirmed HIPAA/BAA posture.
Human approval for all patient/client-facing content.
Prefer admin workflows over clinical workflows.
No Kubernetes, custom auth, broad SaaS platform, or multi-region architecture in V1.

Created: 2026-05-10. Implementation drilldown for Medical, dental, veterinary, and therapy practices. Confidence: medium; validate integrations and compliance requirements with each client.