Compliant GPU Blueprint

This example is designed for healthcare buyers, legal reviewers, compliance stakeholders, and information security teams. It uses DiaScope to walk through a blueprint for compliant GPU-backed AI processing when clinical context leaves a healthcare provider’s current hosting environment. The walkthrough is built around the rented-GPU / self-managed deployment model because that makes responsibility split, provider prerequisites, safeguards, and reviewable evidence legible in one buyer-facing artifact.

Example

Pop out

Included assets

• Open the live story in a new tab
• Download the D2 source
• Download the story sidecar
• Read the example README

What the walkthrough covers

1. Where provider-controlled context leaves the current hosting boundary.
2. Which controller-side decisions and identities must be in place before that handoff.
3. How responsibilities split between controller, implementation partner, and GPU provider.
4. Which legal and evidence prerequisites must exist before a provider can be onboarded.
5. Why the rented-GPU model serves as the compliance blueprint for external AI processing.
6. What evidence must remain reviewable after the response comes back.

Source structure

examples/
  vLLM/
    deployment.d2
    deployment.story.yaml
    README.md

Notes on this example

• examples/vLLM/ is the canonical source for the example.
• The docs site serves public copies from docs-site/public/examples/vllm/.
• The diagram is intentionally dense because the legal and control boundary is the real subject of the walkthrough.
• The story file narrows that density into sequential beats for non-engineering review.

Recreate this workflow with Claude Code or Codex

This example is also a good model for agent-assisted authoring. After installing the DiaScope skill into Claude Code or Codex, you can ask for the same kind of artifact directly:

Use the narrate skill to create a compliance-focused DiaScope walkthrough for our external AI processing flow.

Create a buyer-facing diagram story like the vLLM example, but for our own deployment and control boundaries.

Revise the narration in examples/vLLM/deployment.story.yaml to be shorter and more executive-friendly, then rebuild the HTML.

The agent workflow is documented in AI Agents.