Case Study: Care Transitions Companion

From zero to production in one afternoon

The Pitch

Post-discharge care management is a multi-billion dollar industry. Companies like Wellsky, HealthChannels, and CarePort have built entire businesses around helping patients navigate the transition from hospital to home.

We built a production-ready version in two hours.

Not a prototype. Not a demo. A real product with strategic research, operational playbooks, and working code.

What We Built

Strategic Layer (8 documents)

Before writing any code, we developed a comprehensive understanding:

Incentive Analysis — Unlike payer-provider UM conflicts, incentives are aligned here. Hospitals face HRRP penalties, payers bear readmission costs, patients don't want to go back. The problem isn't alignment — it's execution.
Workflow Mapping — Discharge is designed around the hospital's process (get patient out, check boxes), not the patient's journey (understand what happened, execute the plan, get help). The handoff is a moment; the transition is a period.
Data & Systems — The hospital knows everything at discharge. The problem is it doesn't flow to patients in usable form, and there's no visibility once they leave.
Problem Definition — Three core issues: patient doesn't understand, nobody owns the in-between, nowhere to go but the ED.
Intervention Design — AI care companion: knows the patient's situation, available 24/7, translates clinical info, triages concerns, escalates to humans when needed.
Adoption Strategy — Hospitals are the right entry point. They own discharge, have EHR data, bear readmission penalties. Payer deployment can follow.
Product Specification — Input schema, interaction patterns, triage logic, escalation pathways.
Technical Architecture — Not a skill, a product. Persistent state, multiple user types, scheduled operations, external integrations.

Operational Layer (5 documents)

How it actually runs day-to-day:

Deployment scenarios (who deploys, what needs to be true)
Enrollment workflow (moment-by-moment patient onboarding)
Staff operations (nurse dashboard, escalation handling)
Escalation operations (response times, handoff protocols)
Consent & compliance (HIPAA, documentation, audit trail)

Design Insights (10 key learnings)

Non-obvious insights that shaped the design:

Incentives are aligned — execution is the problem
The patient is invisible post-discharge (no visibility until something goes wrong)
Information exists but isn't accessible (translation and availability problem)
The gap is temporal, not informational (not about better discharge teaching)
Current solutions don't scale ($200-500/patient for 1:1 human support)
SMS beats apps (98% open rates, no download, any phone)
Proactive beats reactive (patients don't know what to ask, don't want to bother anyone)
Triage logic is the core IP (not the conversation — knowing when to escalate)
Escalation pathway is critical (not a phone number — a workflow with context)
Hospitals are the right entry point (own discharge, have data, have incentive)

Production Application

Next.js 14 + TypeScript + Drizzle ORM + Neon Postgres + Claude API

Data Model (11 tables):

Patients, DischargeContexts, Medications, Appointments, WarningSigns
Conversations, Messages, CheckIns, Escalations, TriageDecisions

Core Features:

4-level triage system (REASSURE → MONITOR → ESCALATE → URGENT)
Keyword safety validation that can only override upward (more conservative)
Proactive check-in message generation
Escalation summaries written for nurse audience
Care manager dashboard with patient status
Demo chat interface for testing

Sample Interactions:

"what is the white pill for" → Personalized answer from patient's medication list
"my legs are more swollen today" → Assesses against warning signs, may escalate
"I have chest pain" → Immediate URGENT triage, 911/ED direction

What This Demonstrates

1. Speed That Redefines "Company"

This used to be a company. Entire organizations exist to solve this problem — they raised funding, hired teams, built sales organizations.

We built a production-ready version in an afternoon.

The implication: The unit of "product" has changed. What was once a business is now a feature. What was once a feature is now a prompt.

2. Full-Stack Thinking

This isn't just code generation. The value is in the thinking:

Market research → Why do care transitions fail?
Business strategy → Who's the buyer? What's the business model?
User experience → SMS beats apps. Proactive beats reactive.
Product management → What's the triage logic? What's the escalation pathway?
Technical architecture → Persistent state, scheduled operations, integrations
Implementation → Working code, deployed database, functional demo

All integrated. All in one session.

3. The Real Skill

Anyone can generate code. What's valuable is knowing:

That incentives are aligned (unlike UM) — so this is an execution problem, not a market creation problem
That the problem is temporal not informational — so 24/7 availability matters more than better discharge teaching
That hospitals are the right buyer — because they own the moment, have the data, bear the penalty
That triage logic is the core IP — not the conversation, the judgment

The code is table stakes. The insight is the edge.

4. Proof of the Teaching Flywheel

If we can develop this capability internally — go from "what is care transitions?" to "here's a production app" in two hours — we can develop it for customers.

The bottleneck for AI transformation isn't models. It's talent that knows what to build and why.

We're proof that gap can be closed.

The Fundraising Message

"This used to be a company. Now it's a Tuesday."

The care transitions companion demonstrates what's now possible:

Velocity — Full product development (research → strategy → code) in hours, not months
Scope — Not just technical implementation, but strategic thinking integrated throughout
Quality — Production-ready, not prototype. Real data model, real safety logic, real operational playbooks.

This is what we do for healthcare operations. Utilization management today. Care transitions tomorrow. Whatever the customer needs.

The moat isn't the code. It's knowing what to build.

Artifacts

Repository: /Users/michaelyuan/Code3/research/care-transitions/

care-transitions/
├── context-scaffold.md      # Master overview
├── design-insights.md       # 10 key learnings
├── strategic/               # 8 strategic documents
│   ├── 01-incentives.md
│   ├── 02-workflow.md
│   ├── 03-systems.md
│   ├── 04-problem.md
│   ├── 05-intervention.md
│   ├── 06-adoption.md
│   ├── 07-skill.md
│   └── 08-technical.md
├── operational/             # 5 operational documents
│   ├── 09-deployment-scenarios.md
│   ├── 10-enrollment-workflow.md
│   ├── 11-staff-operations.md
│   ├── 12-escalation-operations.md
│   └── 13-consent-compliance.md
└── app/                     # Production application
    ├── src/
    │   ├── app/             # Next.js app router
    │   ├── db/              # Drizzle schema, seed data
    │   └── lib/             # AI, triage, conversation logic
    ├── package.json
    └── README.md

To run the demo:

cd /Users/michaelyuan/Code3/research/care-transitions/app
npm install
npm run dev
# Open http://localhost:3000