Mindee Bank Statement Extraction

arazzo: 1.0.1
info:
  title: Mindee Bank Statement Extraction
  summary: Enqueue a bank statement with RAG and raw text enabled, poll until processed, then read the extracted transactions and balances.
  description: >-
    Applies the Mindee asynchronous extraction pattern to bank and financial
    statements, which often contain long transaction tables. The workflow
    uploads a statement against a financial extraction model with both the rag
    and raw_text options enabled to improve recall on dense documents, polls
    the shared jobs endpoint until the job is Processed, and fetches the
    inference to read the extracted account fields, transaction lines, and the
    full raw text. 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
workflows:
- workflowId: bank-statement-extraction
  summary: Upload a bank statement with RAG and raw text, then read the extracted fields.
  description: >-
    Sends a bank statement to the extraction enqueue endpoint with rag and
    raw_text enabled, polls the job until processing finishes, and retrieves
    the extracted fields and full text.
  inputs:
    type: object
    required:
    - authorization
    - modelId
    - file
    properties:
      authorization:
        type: string
        description: Mindee API key sent in the Authorization header.
      modelId:
        type: string
        description: UUID of the extraction model trained for bank statements.
      file:
        type: string
        description: The bank statement file to upload as binary form data.
      filename:
        type: string
        description: Optional filename to associate with the uploaded statement.
  steps:
  - stepId: enqueueStatement
    description: >-
      Send the bank statement to the asynchronous extraction queue with rag and
      raw_text enabled to maximize recall on dense transaction tables.
    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.modelId
        file: $inputs.file
        filename: $inputs.filename
        rag: true
        raw_text: true
    successCriteria:
    - condition: $statusCode == 202
    outputs:
      jobId: $response.body#/job/id
      status: $response.body#/job/status
  - stepId: pollJob
    description: >-
      Poll the shared jobs endpoint until the bank statement 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.enqueueStatement.outputs.jobId
    - name: redirect
      in: query
      value: false
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      status: $response.body#/job/status
    onSuccess:
    - name: jobProcessed
      type: goto
      stepId: getResult
      criteria:
      - context: $response.body
        condition: $.job.status == "Processed"
        type: jsonpath
    - name: jobPending
      type: goto
      stepId: pollJob
      criteria:
      - context: $response.body
        condition: $.job.status == "Processing"
        type: jsonpath
  - stepId: getResult
    description: >-
      Retrieve the completed extraction inference and read the structured
      account and transaction fields along with the full raw text.
    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.enqueueStatement.outputs.jobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      inferenceId: $response.body#/inference/id
      fields: $response.body#/inference/result/fields
      rawText: $response.body#/inference/result/raw_text
  outputs:
    jobId: $steps.enqueueStatement.outputs.jobId
    fields: $steps.getResult.outputs.fields
    rawText: $steps.getResult.outputs.rawText