[bmdpat]

§ 001 / AGENTFLOW

Watch two AI agents coordinate care - live.

Pick a patient scenario. Watch all 5 pipeline steps in real time: barrier detection, 1 tool call, 1 context handoff, resource matching, then exactly what the care coordinator does next.

Synthetic patient data — no PHI processed in this demo

§ 002 / SCENARIO INPUT

In production, scenario arrives via:

SMS / Textpatient texts care-line shortcode
Voice / IVRcall transcript → triage input
CHW Intakecoordinator enters on patient's behalf
EHR TriggerHCC code or chief complaint from EMR

§ 004 / HOW IT WORKS

skill.md

Each agent has a system prompt defining its identity, triggers, and output schema. Visible in every panel above.

context handoff

Agents share no chat history. They pass typed JSON — a contract that makes the next agent deterministic.

tool calls

Trigger fires → structured API call → result comes back as data. The next agent reasons on it, not a string.

human in the loop

The CHW panel shows exactly where a person takes over. AI surfaces the plan — the coordinator owns the action.

§ 005 / STACK

this demo

  • Frontend
    Next.js 15App Router + streaming API route
  • LLM
    NVIDIA NIMLlama 3.3 70B via OpenAI-compatible API / 30s timeout / 1024 max tokens
  • Streaming
    ReadableStreamnewline-delimited JSON events, no WebSocket
  • Deploy
    Verceledge-optimized, instant preview

production-ready

  • Agents
    Python + LangGraphstateful multi-agent orchestration
  • API
    FastAPIasync, typed, HIPAA-scoped endpoints
  • Data
    FHIR R4 + PostgreSQLstructured patient records + relational case data
  • Queue
    Redis + Celeryasync task processing, retries, dead-letter queue
  • Infra
    AWS GovCloud / Azure GovHIPAA BAA · SOC 2 · FedRAMP eligible

Get The One-Person Holdco (free PDF)

How one human plus twenty-two AI agents runs a seven-pillar portfolio with no employees.