Triage Park

A conversational FHIR triage assistant. A patient answers a short intake interview; an LLM-backed agent reads their FHIR record, retrieves matching triage guidelines by vector search, and produces a clinician handoff — a triage level (self-care / see-GP / urgent-care / ED) with a cited rationale. Every interview and its outcome are written back to FHIR, and a clinician reviews them in a worklist.

Built for the InterSystems Programming Contest: AI Agents for FHIR.

Quickstart

git clone https://github.com/eungi-hong/central-park.git
cd central-park
cp .env.example .env          # then set OPENAI_API_KEY=sk-...
docker compose up --build      # IRIS cold start ~90s

Then open:

Once seeding finishes, the console shows three example cases (self-care, see-GP, and an emergency) — no interview needed. Then open /intake, run the demo patient demo-patient-1 through the chest-tightness scenario, and watch a new case appear live in the worklist.

What it does

Two front doors, one FHIR backend.

• Patient intake (/intake) — a first-person interview: chief complaint, onset, a 1–10 severity scale, a symptom checklist, history, and self-treatment. Answers are saved to FHIR as a QuestionnaireResponse; the patient gets a plain-language next step.
• Clinician console (/) — a worklist of triaged cases, newest first, with urgent cases flagged. Opening a case shows the patient’s standing record, the interview transcript, the agent’s assessment and cited guidelines, and an Acknowledge action for escalated cases (written back to FHIR).

Architecture

            browser
   ┌───────────┴────────────┐
   /intake (patient)   /  (clinician)
   └───────────┬────────────┘
        central-park-ui  (React SPA + nginx :8501)
        ├─ /fhir/* → IRIS FHIR R4   (Basic auth injected server-side)
        └─ /api/*  → agent
                       │
   ┌───────────────────┼─────────────────────────────────┐
   │ central-park-iris (IRIS for Health)                 │
   │   FHIR R4 endpoint  ·  REST dispatch (/triage …)    │
   │   Interoperability production (REST inbox → agent   │
   │     business operation → Ens alerts on escalation)  │
   │   %Embedding.OpenAI config  ·  VECTOR_COSINE search │
   └───────────────────┼─────────────────────────────────┘
                        │ POST /run · /interview
        central-park-agent (FastAPI + LangGraph)
          gather_context → retrieve_guidelines →
          red-flag gate → reason → escalate
                        │
                  OpenAI (chat)

Three services, one external dependency (OpenAI). The agent sends raw text to IRIS, which computes embeddings server-side via %Embedding.OpenAI — the platform’s intended AI Hub pattern. A LangGraph state machine wraps the five-node flow; a deterministic red-flag gate can only ever escalate, never downgrade, so a missed keyword can’t lower a triage level below a matched emergency phrase.

The agent runs as a sidecar rather than Embedded Python: the image’s ARM64 Embedded Python build is unstable (Callin SEGVs, broken _uuid). The IRIS production graph stays first-class — every triage is an Ens.MessageHeader visible in Visual Trace.

InterSystems features used

FHIR resources written

Persisting the narrative on the ServiceRequest is what lets the console reconstruct a full past case from FHIR alone — no LLM re-run.

Project layout

.
├─ ui/                      # React SPA — clinician console (/) + patient intake (/intake)
├─ agent/                   # Dockerfile for the FastAPI + LangGraph sidecar
├─ src/
│  ├─ python/central_park/  # Agent: LangGraph graph, tools (FHIR · vector · escalate), seeding
│  └─ cls/CentralPark/      # ObjectScript: FHIR install, REST dispatch, interop production
├─ iris-config/             # Boot script + demo seed bundles (patients, questionnaire)
├─ web/                     # Static assets served by IRIS
├─ Dockerfile               # IRIS for Health image: namespace, FHIR R4 endpoint, classes
├─ docker-compose.yml       # iris + agent + ui  (+ optional ollama profile)
└─ module.xml               # OpenExchange / IPM manifest

Internal identifiers keep the original central-park / CentralPark / /centralpark names (package, classes, REST path, containers); “Triage Park” is the product/display name.

Demo data

Seeded automatically at agent startup: the agent waits for the FHIR endpoint, then loads the bundles in iris-config/seed/ (and the guideline corpus). Every seed is idempotent, so it re-runs safely on each restart and a single docker compose up needs no manual seed step.

Marcus is deliberately cardiac-risk-loaded so a chest-tightness scenario exercises real reasoning over his FHIR record.

Verify

curl http://localhost:8001/health                                   # agent
curl -u _SYSTEM:SYS http://localhost:52773/centralpark/health        # IRIS
Vector search (no LLM cost)
curl -X POST http://localhost:52773/centralpark/vector/search -u _SYSTEM:SYS 

-H 'Content-Type: application/json' -d '{"query":"chest tightness on exertion","k":3}'
Direct triage (chat LLM); writes a Communication when urgent
curl -X POST http://localhost:52773/centralpark/triage -u _SYSTEM:SYS 

-H 'Content-Type: application/json' 

-d '{"patient_id":"demo-patient-1","message":"My chest feels tight when I walk upstairs."}'

Visual Trace shows every triage as an Ens message; the Embedding config lives under System Administration → Configuration → Connectivity → Embedding Configurations.

Configuration

Defaults are baked into docker-compose.yml; .env overrides them.

Switch chat to Anthropic (CP_LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) or offline Ollama (docker compose --profile ollama up --build); OpenAI is still required for embeddings. Run docker compose restart agent after changing .env.

Notes

• Iterate — Python under src/python/ is bind-mounted: docker compose restart agent. ObjectScript and boot config need docker compose up --build.
• Seeding runs automatically at agent startup and is idempotent. If the worklist ever looks empty (e.g. IRIS was unusually slow to start), docker compose restart agent re-runs the seed safely.
• Data persists in the iris-data volume across restarts. docker compose down -v wipes it and re-seeds on next boot.
• License — MIT, see LICENSE.