Healthcare Multi-Agent System

Applied Agentic AI Program · Interview Kickstart · Spring 2026
Lamonte Smith

A conversational multi-agent AI system built in n8n that orchestrates three specialized healthcare agents — Doctor Booking, Hospital & ER, and Diagnostics — through a central Coordinator Agent. Powered by GPT-4o and GPT-4o-mini, backed by five live Google Sheets datasets covering 3,400+ hospitals, 28 doctors across 12 specialties, and 8 diagnostic test types.

---

System Architecture

User (Chat Interface)
        │
        ▼
┌────────────────────────────────────────┐
│        Healthcare Coordinator          │
│           (GPT-4o)                     │
│  • Intent classification               │
│  • Multi-intent routing                │
│  • Window Buffer Memory (10 turns)     │
│  • Seamless agent orchestration        │
└────┬──────────────┬────────────────────┘
     │              │              │
     ▼              ▼              ▼
┌─────────┐  ┌───────────┐  ┌────────────┐
│ Doctor  │  │ Hospital  │  │Diagnostics │
│ Booking │  │  & ER     │  │   Agent    │
│  Agent  │  │  Agent    │  │(GPT-4o-mini│
│(GPT-4o- │  │(GPT-4o-   │  │            │
│  mini)  │  │  mini)    │  │            │
└────┬────┘  └─────┬─────┘  └─────┬──────┘
     │             │              │
  ┌──┴──┐    ┌─────┴──────┐  ┌───┴──────────┐
  │ 📋  │    │     🏥     │  │      🔬      │
  │ Dr  │    │  Hospital  │  │  Lab Tests   │
  │Info │    │  General   │  │   Dataset    │
  │ 📅  │    │    Info    │  │  ✉️ Gmail    │
  │ Dr  │    │  🚨 ER &   │  │  (prep docs) │
  │Avail│    │ Emergency  │  │              │
  └─────┘    └────────────┘  └──────────────┘

---

Agents

Healthcare Coordinator Agent

• Model: GPT-4o (coordinator-tier)
• Role: Primary user interface — classifies intent and routes to specialist agents
• Memory: Window Buffer Memory — 10-turn conversation context
• Capability: Multi-intent routing — can dispatch to multiple agents simultaneously and synthesize a unified response
• Routing logic: Doctor queries → Doctor Booking Agent · Hospital/ER queries → Hospital & ER Agent · Diagnostic queries → Diagnostics Agent

Doctor Booking Agent

• Model: GPT-4o-mini
• Datasets: Doctor Info · Doctor Availability
• Capabilities:
  • Find doctors by specialty across 12 medical specialties
  • Check real-time appointment availability (March 1–7, 8:00 AM–5:30 PM, 30-min slots)
  • Book and confirm appointments
  • Provide doctor contact information
• Specialties: Nephrology · Orthopedics · Ophthalmology · Cardiology · Radiology · Gynecology · Pediatrics · Psychiatry · Pulmonology · Endocrinology · Dermatology · Oncology

Hospital & ER Agent

• Model: GPT-4o-mini
• Datasets: Hospital General Info (3,401 hospitals) · Hospital Emergency Data (32 ER locations)
• Capabilities:
  • Find hospitals by name, city, or state
  • Compare quality ratings (1–5 stars) across mortality, safety, readmission, and patient experience metrics
  • Locate emergency rooms with ambulance availability by ZIP code
  • Report on CMS quality metrics
• ER Coverage: NYC · DC · Atlanta · Miami · Chicago · New Orleans · Austin · Phoenix · LA · SF Bay Area · Seattle

Diagnostics Agent

• Model: GPT-4o-mini
• Datasets: Hospital Lab Tests · Health Packages
• Capabilities:
  • Provide preparation instructions for 8 diagnostic test types
  • Surface health package options (Cancer Screening, Diabetes Management, Heart Care, Orthopedic, Women's Wellness, Full Body Checkup)
  • Email test prep instructions to patient via Gmail integration
  • Identify hospitals offering specific tests

---

Datasets (5 Google Sheets)

Dataset	Records	Coverage
Doctor Info	28 doctors	12 specialties
Doctor Availability	Appointment slots	March 1–7, 30-min intervals
Hospital General Info	3,401 hospitals	Nationwide — CMS quality metrics
Hospital Emergency Data	32 ER locations	13 major metro ZIP codes
Hospital Lab Tests	8 test types	Prep instructions + health packages

---

Tech Stack

Models: GPT-4o (coordinator) · GPT-4o-mini (specialist agents)
Integrations: Google Sheets (5 datasets) · Gmail (email delivery)
Patterns: Coordinator-Worker · Multi-Intent Routing · Tool-as-Agent · Window Buffer Memory

---

Workflow Features

• Seamless UX — internal agent names never exposed to the user
• Multi-intent handling — single query can trigger multiple agents simultaneously
• Persistent memory — 10-turn context window per session on the coordinator
• Graceful fallback — off-topic queries politely redirected
• Email integration — diagnostic prep instructions deliverable via Gmail

---

Sample Queries

"Find me a cardiologist available Monday morning"
"Which hospitals near ZIP 60601 have 5-star ratings and ambulance availability?"
"What do I need to do to prepare for an MRI?"
"Book an appointment with a pediatrician and email me the confirmation"
"Compare the top-rated hospitals in Michigan"

---

Academic Context

Field	Detail
Program	Applied Agentic AI
Institution	Interview Kickstart
Semester	Spring 2026
Assignment	Week 3 — Multi-Agent Orchestration

---

Repository Structure

healthcare-multi-agent/
├── workflow/
│   └── Healthcare_Multi_Agent_System_Combined.json   # n8n export
├── docs/
│   └── agent-routing.md                              # Routing logic detail
├── SECURITY.md
└── README.md

---

Security Notes

• All Google Sheets credentials managed via n8n credential store — never hardcoded
• Gmail integration uses OAuth2 — no plaintext passwords
• Dataset IDs are referenced by n8n internal ID — not exposed in workflow logic
• See SECURITY.md for full policy

---

License

MIT License — see LICENSE

---

_{Built by Lamonte Smith · Applied Agentic AI · Interview Kickstart}