Mindee OCR Then Extract

arazzo: 1.0.1
info:
  title: Mindee OCR Then Extract
  summary: Run OCR over a document to capture its raw text, then extract structured fields from the same file, reading both outputs.
  description: >-
    A two-product enrichment pattern. The workflow first runs the OCR utility
    to capture the full per-page text of a document, waits for that job and
    reads the pages, then enqueues the same file for extraction against the
    supplied extraction model, polls the extraction job, and reads the parsed
    fields. The combination yields both a faithful text transcription and a
    structured field set from one source file. 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
sourceDescriptions:
- name: extractionApi
  url: ../openapi/mindee-extraction-api-openapi.yml
  type: openapi
- name: jobsApi
  url: ../openapi/mindee-jobs-api-openapi.yml
  type: openapi
- name: ocrApi
  url: ../openapi/mindee-ocr-api-openapi.yml
  type: openapi
workflows:
- workflowId: ocr-then-extract
  summary: OCR a document for raw text, then extract structured fields from it.
  description: >-
    Runs OCR to capture per-page text, then enqueues the same file for
    extraction, polling each job to completion and reading the pages and the
    extracted fields.
  inputs:
    type: object
    required:
    - authorization
    - ocrModelId
    - extractionModelId
    - file
    properties:
      authorization:
        type: string
        description: Mindee API key sent in the Authorization header.
      ocrModelId:
        type: string
        description: UUID of the OCR utility model.
      extractionModelId:
        type: string
        description: UUID of the extraction model to apply after OCR.
      file:
        type: string
        description: The document file to upload as binary form data.
      filename:
        type: string
        description: Optional filename to associate with the uploaded document.
  steps:
  - stepId: enqueueOcr
    description: >-
      Send the document to the asynchronous OCR queue to capture its full text.
    operationId: Enqueue_OCR_Product_Inference_v2_products_ocr_enqueue_post
    parameters:
    - name: Authorization
      in: header
      value: $inputs.authorization
    requestBody:
      contentType: multipart/form-data
      payload:
        model_id: $inputs.ocrModelId
        file: $inputs.file
        filename: $inputs.filename
    successCriteria:
    - condition: $statusCode == 202
    outputs:
      ocrJobId: $response.body#/job/id
  - stepId: pollOcr
    description: >-
      Poll the shared jobs endpoint until the OCR job reports Processed or
      Failed.
    operationId: Get_Job_Status_v2_jobs__job_id__get
    parameters:
    - name: Authorization
      in: header
      value: $inputs.authorization
    - name: job_id
      in: path
      value: $steps.enqueueOcr.outputs.ocrJobId
    - name: redirect
      in: query
      value: false
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      status: $response.body#/job/status
    onSuccess:
    - name: ocrProcessed
      type: goto
      stepId: getOcr
      criteria:
      - context: $response.body
        condition: $.job.status == "Processed"
        type: jsonpath
    - name: ocrPending
      type: goto
      stepId: pollOcr
      criteria:
      - context: $response.body
        condition: $.job.status == "Processing"
        type: jsonpath
  - stepId: getOcr
    description: >-
      Read the recognized per-page text from the completed OCR inference.
    operationId: Get_OCR_Product_Result_v2_products_ocr_results__inference_id__get
    parameters:
    - name: Authorization
      in: header
      value: $inputs.authorization
    - name: inference_id
      in: path
      value: $steps.enqueueOcr.outputs.ocrJobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      pages: $response.body#/inference/result/pages
  - stepId: enqueueExtraction
    description: >-
      Send the same file to the extraction queue against the chosen extraction
      model to read its structured fields.
    operationId: Enqueue_Extraction_Product_Inference_v2_products_extraction_enqueue_post
    parameters:
    - name: Authorization
      in: header
      value: $inputs.authorization
    requestBody:
      contentType: multipart/form-data
      payload:
        model_id: $inputs.extractionModelId
        file: $inputs.file
        filename: $inputs.filename
    successCriteria:
    - condition: $statusCode == 202
    outputs:
      extractionJobId: $response.body#/job/id
  - stepId: pollExtraction
    description: >-
      Poll the shared jobs endpoint until the extraction job reports Processed
      or Failed.
    operationId: Get_Job_Status_v2_jobs__job_id__get
    parameters:
    - name: Authorization
      in: header
      value: $inputs.authorization
    - name: job_id
      in: path
      value: $steps.enqueueExtraction.outputs.extractionJobId
    - name: redirect
      in: query
      value: false
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      status: $response.body#/job/status
    onSuccess:
    - name: extractionProcessed
      type: goto
      stepId: getExtraction
      criteria:
      - context: $response.body
        condition: $.job.status == "Processed"
        type: jsonpath
    - name: extractionPending
      type: goto
      stepId: pollExtraction
      criteria:
      - context: $response.body
        condition: $.job.status == "Processing"
        type: jsonpath
  - stepId: getExtraction
    description: >-
      Retrieve the completed extraction inference and read the structured
      fields parsed from the document.
    operationId: Get_Extraction_Product_Result_v2_products_extraction_results__inference_id__get
    parameters:
    - name: Authorization
      in: header
      value: $inputs.authorization
    - name: inference_id
      in: path
      value: $steps.enqueueExtraction.outputs.extractionJobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      inferenceId: $response.body#/inference/id
      fields: $response.body#/inference/result/fields
  outputs:
    pages: $steps.getOcr.outputs.pages
    fields: $steps.getExtraction.outputs.fields