Amazon Neptune ML Create and Verify Inference Endpoint

arazzo: 1.0.1
info:
  title: Amazon Neptune ML Create and Verify Inference Endpoint
  summary: Create an ML inference endpoint from a trained model and poll it until it is in service.
  description: >-
    Stands up a Neptune ML inference endpoint from a completed model training
    job. The workflow creates the endpoint, polls its status on a loop until it
    reaches InService, and then lists the active inference endpoints to confirm
    the new endpoint is registered. The poll loop uses a retry delay to handle the
    Creating state. Every step spells out its request inline so the flow can be
    read and executed without opening the underlying OpenAPI description.
  version: 1.0.0
  x-realizes-capability-ids:
  - BC-610.60
  x-capability-derivation:
    method: 'deterministic join: sourceDescriptions -> per-tag OpenAPI -> tag/capability edge. No classification at this step.'
    min_confidence: 0.7
    sources:
    - capability_id: BC-610.60
      capability_name: Artificial Intelligence Management
      spec: amazon-neptune-inference-endpoints-api-openapi.yml
      confidence: 0.8
    model: Turbo EA Capabilities by Vincent Verdet — Turbo EA, https://github.com/vincentmakes/turbo-ea-capabilities, CC BY 4.0
sourceDescriptions:
- name: inferenceEndpointsApi
  url: ../openapi/amazon-neptune-inference-endpoints-api-openapi.yml
  type: openapi
workflows:
- workflowId: ml-create-inference-endpoint
  summary: Create an inference endpoint, poll until InService, and confirm it is listed.
  description: >-
    Creates an ML inference endpoint from a training job, polls until it is
    InService, and lists endpoints to confirm registration.
  inputs:
    type: object
    required:
    - mlModelTrainingJobId
    properties:
      mlModelTrainingJobId:
        type: string
        description: Job ID from a completed model training job.
      instanceType:
        type: string
        description: ML instance type for the inference endpoint.
      instanceCount:
        type: integer
        description: Minimum number of EC2 instances to deploy.
  steps:
  - stepId: createEndpoint
    description: >-
      Create an inference endpoint backed by the trained model and capture the
      endpoint id.
    operationId: createInferenceEndpoint
    requestBody:
      contentType: application/json
      payload:
        mlModelTrainingJobId: $inputs.mlModelTrainingJobId
        instanceType: $inputs.instanceType
        instanceCount: $inputs.instanceCount
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      endpointId: $response.body#/id
  - stepId: pollEndpoint
    description: >-
      Poll the endpoint status. Retry while it is not yet InService and continue
      once it reaches InService.
    operationId: getInferenceEndpointStatus
    parameters:
    - name: id
      in: path
      value: $steps.createEndpoint.outputs.endpointId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      endpointStatus: $response.body#/status
    onSuccess:
    - name: endpointCreating
      type: retry
      retryAfter: 60
      retryLimit: 120
      criteria:
      - context: $response.body
        condition: $.status != "InService"
        type: jsonpath
    - name: endpointReady
      type: goto
      stepId: listEndpoints
      criteria:
      - context: $response.body
        condition: $.status == "InService"
        type: jsonpath
  - stepId: listEndpoints
    description: >-
      List the active inference endpoints to confirm the newly created endpoint
      is registered.
    operationId: listInferenceEndpoints
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      endpointIds: $response.body#/ids
  outputs:
    endpointId: $steps.createEndpoint.outputs.endpointId
    endpointStatus: $steps.pollEndpoint.outputs.endpointStatus
    endpointIds: $steps.listEndpoints.outputs.endpointIds