Consultation Agent

Overview

Layer 4 on the Mental Model. The Consultation agent is a single streaming path behind FastAPI: verify Clerk JWT, then stream a frontier-model reply over SSE for the clinician form.

Behavior

• Auth before inference. No consultation stream without a verified JWT (ClerkHTTPBearer + JWKS).
• Server-side model only. The UI posts notes; FastAPI owns the OpenAI call and SSE deltas.
• Fixed three-section reply. System prompt asks for visit summary, next steps for the doctor, and a patient-friendly draft email.
• Stream then render. Tokens arrive as SSE; the client renders markdown as they land.
• Prompt Chaining. Visit fields → structured sections in one streamed completion. See Building effective agents.

Consultation flow

One signed-in clinician consultation. Diagram key: ✦ AI agent · second line = Anthropic workflow · purple = actor · white = service.

---
config:
  theme: default
  themeVariables:
    background: "#ffffff"
    fontSize: "14px"
---
flowchart TB
    person[Clinician]
    form[Product form]
    api["POST /api/consultation"]
    clerk[Clerk JWKS verify]
    consultation["✦ Consultation
Prompt Chaining"]
    model["OpenAI stream
gpt-5-nano"]
    sse[SSE deltas]
    md[ReactMarkdown reply]

    person -->|submit notes| form -->|Bearer JWT| api
    api --> clerk --> consultation
    consultation --> model --> sse --> md --> person

    classDef agent fill:#e8f0ee,stroke:#0f3d36,stroke-width:2px,color:#0d0d0d
    classDef service fill:#fafaf8,stroke:#c8c8c2,stroke-width:1px,color:#0d0d0d
    classDef actor fill:#f3eef8,stroke:#6b4f9a,stroke-width:2px,color:#0d0d0d
    class consultation agent
    class form,api,clerk,model,sse,md service
    class person actor

Route: POST /api/consultation. Delivery: SSE to the form page. Research deployment only — not a licensed clinical product.