✨ add support for workflow polling (#185)

Author: sebastianMindee

.github/workflows/_static-analysis.yml

@@ -5,6 +5,7 @@ name: Static Analysis
 
 on:
   workflow_call:
+  workflow_dispatch:
 
 jobs:
   linting:
@@ -18,7 +19,7 @@ jobs:
     - name: set up Ruby
       uses: ruby/setup-ruby@v1
       with:
-        ruby-version: "3.0"
+        ruby-version: "3.0.0"
         bundler-cache: true
 
     - name: Analyse the code with Rubocop

.rubocop.yml

@@ -10,7 +10,7 @@ AllCops:
     - local_test/*
     - Steepfile
 
-  TargetRubyVersion: 3.0
+  TargetRubyVersion: 3.0.0
   SuggestExtensions: false
 
 Gemspec/DevelopmentDependencies:

docs/code_samples/workflow_execution.txt

@@ -3,7 +3,7 @@
 # gem install mindee
 #
 
-require_relative 'mindee'
+require 'mindee'
 
 workflow_id = 'workflow-id'
 

docs/code_samples/workflow_polling.txt

@@ -0,0 +1,36 @@
+#
+# Install the Ruby client library by running:
+# gem install mindee
+#
+
+require 'mindee'
+
+workflow_id = 'workflow-id'
+
+# Init a new client
+mindee_client = Mindee::Client.new
+
+# Load a file from disk
+input_source = mindee_client.source_from_path('path/to/my/file.ext')
+
+# Initialize a custom endpoint for this product
+custom_endpoint = mindee_client.create_endpoint(
+  account_name: 'my-account',
+  endpoint_name: 'my-endpoint',
+  version: 'my-version'
+)
+
+# Parse the file
+result = mindee_client.parse(
+  input_source,
+  Mindee::Product::Universal::Universal,
+  endpoint: custom_endpoint,
+  options: {
+    rag: true,
+    workflow_id: workflow_id
+  }
+)
+
+# Print a full summary of the parsed data in RST format
+puts result.document
+

lib/mindee/client.rb

@@ -54,8 +54,8 @@ def initialize(params: {})
   # @!attribute delay_sec [Numeric] Delay between polling attempts. Defaults to 1.5.
   # @!attribute max_retries [Integer] Maximum number of retries. Defaults to 80.
   class ParseOptions
-    attr_accessor :all_words, :full_text, :close_file, :page_options, :cropper,
-                  :initial_delay_sec, :delay_sec, :max_retries
+    attr_accessor :all_words, :full_text, :close_file, :page_options, :cropper, :rag,
+                  :workflow_id, :initial_delay_sec, :delay_sec, :max_retries
 
     def initialize(params: {})
       params = params.transform_keys(&:to_sym)
@@ -66,6 +66,8 @@ def initialize(params: {})
       raw_page_options = PageOptions.new(params: raw_page_options) unless raw_page_options.is_a?(PageOptions)
       @page_options = raw_page_options
       @cropper = params.fetch(:cropper, false)
+      @rag = params.fetch(:rag, false)
+      @workflow_id = params.fetch(:workflow_id, nil)
       @initial_delay_sec = params.fetch(:initial_delay_sec, 2)
       @delay_sec = params.fetch(:delay_sec, 1.5)
       @max_retries = params.fetch(:max_retries, 80)
@@ -176,13 +178,10 @@ def parse_sync(input_source, product_class, endpoint, options)
 
       prediction, raw_http = endpoint.predict(
         input_source,
-        options.all_words,
-        options.full_text,
-        options.close_file,
-        options.cropper
+        options
       )
 
-      Mindee::Parsing::Common::ApiResponse.new(product_class, prediction, raw_http)
+      Mindee::Parsing::Common::ApiResponse.new(product_class, prediction, raw_http.to_s)
     end
 
     # Enqueue a document for async parsing
@@ -207,6 +206,8 @@ def parse_sync(input_source, product_class, endpoint, options)
     #       - `:on_min_pages` [Integer] Apply the operation only if the document has at least this many pages.
     #   * `:cropper` [bool] Whether to include cropper results for each page.
     #       This performs a cropping operation on the server and will increase response time.
+    #   * `:rag` [bool] Whether to enable Retrieval-Augmented Generation. Only works if a Workflow ID is provided.
+    #   * `:workflow_id` [String, nil] ID of the workflow to use.
     # @param endpoint [Mindee::HTTP::Endpoint] Endpoint of the API.
     # @return [Mindee::Parsing::Common::ApiResponse]
     def enqueue(input_source, product_class, endpoint: nil, options: {})
@@ -216,12 +217,9 @@ def enqueue(input_source, product_class, endpoint: nil, options: {})
 
       prediction, raw_http = endpoint.predict_async(
         input_source,
-        opts.all_words,
-        opts.full_text,
-        opts.close_file,
-        opts.cropper
+        opts
       )
-      Mindee::Parsing::Common::ApiResponse.new(product_class, prediction, raw_http)
+      Mindee::Parsing::Common::ApiResponse.new(product_class, prediction, raw_http.to_json)
     end
 
     # Parses a queued document
@@ -236,7 +234,7 @@ def parse_queued(job_id, product_class, endpoint: nil)
       endpoint = initialize_endpoint(product_class) if endpoint.nil?
       logger.debug("Fetching queued document as '#{endpoint.url_root}'")
       prediction, raw_http = endpoint.parse_async(job_id)
-      Mindee::Parsing::Common::ApiResponse.new(product_class, prediction, raw_http)
+      Mindee::Parsing::Common::ApiResponse.new(product_class, prediction, raw_http.to_json)
     end
 
     # Enqueue a document for async parsing and automatically try to retrieve it
@@ -261,6 +259,8 @@ def parse_queued(job_id, product_class, endpoint: nil)
     #       - `:on_min_pages` [Integer] Apply the operation only if the document has at least this many pages.
     #   * `:cropper` [bool, nil] Whether to include cropper results for each page.
     #       This performs a cropping operation on the server and will increase response time.
+    #   * `:rag` [bool] Whether to enable Retrieval-Augmented Generation. Only works if a Workflow ID is provided.
+    #   * `:workflow_id` [String, nil] ID of the workflow to use.
     #   * `:initial_delay_sec` [Numeric] Initial delay before polling. Defaults to 2.
     #   * `:delay_sec` [Numeric] Delay between polling attempts. Defaults to 1.5.
     #   * `:max_retries` [Integer] Maximum number of retries. Defaults to 80.

lib/mindee/http/endpoint.rb

@@ -34,6 +34,8 @@ class Endpoint
       attr_reader :request_timeout
       # @return [String]
       attr_reader :url_root
+      # @return [String]
+      attr_reader :base_url
 
       def initialize(owner, url_name, version, api_key: '')
         @owner = owner
@@ -44,25 +46,19 @@ def initialize(owner, url_name, version, api_key: '')
           logger.debug('API key set from environment')
         end
         @api_key = api_key.nil? || api_key.empty? ? ENV.fetch(API_KEY_ENV_NAME, API_KEY_DEFAULT) : api_key
-        base_url = ENV.fetch(BASE_URL_ENV_NAME, BASE_URL_DEFAULT)
-        @url_root = "#{base_url.chomp('/')}/products/#{@owner}/#{@url_name}/v#{@version}"
+        @base_url = ENV.fetch(BASE_URL_ENV_NAME, BASE_URL_DEFAULT).chomp('/')
+        @url_root = "#{@base_url}/products/#{@owner}/#{@url_name}/v#{@version}"
       end
 
       # Call the prediction API.
       # @param input_source [Mindee::Input::Source::LocalInputSource, Mindee::Input::Source::URLInputSource]
-      # @param all_words [bool] Whether the full word extraction needs to be performed
-      # @param full_text [bool] Whether to include the full OCR text response in compatible APIs
-      # @param close_file [bool] Whether the file will be closed after reading
-      # @param cropper [bool] Whether a cropping operation will be applied
+      # @param opts [ParseOptions] Parse options.
       # @return [Array]
-      def predict(input_source, all_words, full_text, close_file, cropper)
+      def predict(input_source, opts)
         check_api_key
         response = predict_req_post(
           input_source,
-          all_words: all_words,
-          full_text: full_text,
-          close_file: close_file,
-          cropper: cropper
+          opts
         )
         if !response.nil? && response.respond_to?(:body)
           hashed_response = JSON.parse(response.body, object_class: Hash)
@@ -76,14 +72,11 @@ def predict(input_source, all_words, full_text, close_file, cropper)
 
       # Call the prediction API.
       # @param input_source [Mindee::Input::Source::LocalInputSource, Mindee::Input::Source::URLInputSource]
-      # @param all_words [bool] Whether the full word extraction needs to be performed
-      # @param full_text [bool] Whether to include the full OCR text response in compatible APIs.
-      # @param close_file [bool] Whether the file will be closed after reading
-      # @param cropper [bool] Whether a cropping operation will be applied
+      # @param opts [ParseOptions, Hash] Parse options.
       # @return [Array]
-      def predict_async(input_source, all_words, full_text, close_file, cropper)
+      def predict_async(input_source, opts)
         check_api_key
-        response = document_queue_req_get(input_source, all_words, full_text, close_file, cropper)
+        response = document_queue_req_post(input_source, opts)
         if !response.nil? && response.respond_to?(:body)
           hashed_response = JSON.parse(response.body, object_class: Hash)
           return [hashed_response, response.body] if ResponseValidation.valid_async_response?(response)
@@ -100,7 +93,7 @@ def predict_async(input_source, all_words, full_text, close_file, cropper)
       # @return [Array]
       def parse_async(job_id)
         check_api_key
-        response = document_queue_req(job_id)
+        response = document_queue_req_get(job_id)
         hashed_response = JSON.parse(response.body, object_class: Hash)
         return [hashed_response, response.body] if ResponseValidation.valid_async_response?(response)
 
@@ -112,17 +105,14 @@ def parse_async(job_id)
       private
 
       # @param input_source [Mindee::Input::Source::LocalInputSource, Mindee::Input::Source::URLInputSource]
-      # @param all_words [bool] Whether the full word extraction needs to be performed
-      # @param full_text [bool] Whether to include the full OCR text response in compatible APIs.
-      # @param close_file [bool] Whether the file will be closed after reading
-      # @param cropper [bool] Whether a cropping operation will be applied
+      # @param opts [ParseOptions] Parse options.
       # @return [Net::HTTPResponse, nil]
-      def predict_req_post(input_source, all_words: false, full_text: false, close_file: true, cropper: false)
+      def predict_req_post(input_source, opts)
         uri = URI("#{@url_root}/predict")
 
         params = {} # : Hash[Symbol | String, untyped]
-        params[:cropper] = 'true' if cropper
-        params[:full_text_ocr] = 'true' if full_text
+        params[:cropper] = 'true' if opts.cropper
+        params[:full_text_ocr] = 'true' if opts.full_text
         uri.query = URI.encode_www_form(params)
 
         headers = {
@@ -131,32 +121,33 @@ def predict_req_post(input_source, all_words: false, full_text: false, close_fil
         }
         req = Net::HTTP::Post.new(uri, headers)
         form_data = if input_source.is_a?(Mindee::Input::Source::URLInputSource)
-                      [['document', input_source.url]]
+                      [['document', input_source.url]] # : Array[untyped]
                     else
-                      [input_source.read_contents(close: close_file)]
+                      [input_source.read_contents(close: opts.close_file)] # : Array[untyped]
                     end
-        form_data.push ['include_mvision', 'true'] if all_words
+        form_data.push ['include_mvision', 'true'] if opts.all_words
 
         req.set_form(form_data, 'multipart/form-data')
-        response = nil
         Net::HTTP.start(uri.hostname, uri.port, use_ssl: true, read_timeout: @request_timeout) do |http|
-          response = http.request(req)
+          return http.request(req)
         end
-        response
+        raise Mindee::Errors::MindeeError, 'Could not resolve server response.'
       end
 
       # @param input_source [Mindee::Input::Source::LocalInputSource, Mindee::Input::Source::URLInputSource]
-      # @param all_words [bool] Whether the full word extraction needs to be performed
-      # @param full_text [bool] Whether to include the full OCR text response in compatible APIs.
-      # @param close_file [bool] Whether the file will be closed after reading
-      # @param cropper [bool] Whether a cropping operation will be applied
-      # @return [Net::HTTPResponse, nil]
-      def document_queue_req_get(input_source, all_words, full_text, close_file, cropper)
-        uri = URI("#{@url_root}/predict_async")
+      # @param opts [ParseOptions] Parse options.
+      # @return [Net::HTTPResponse]
+      def document_queue_req_post(input_source, opts)
+        uri = if opts.workflow_id
+                URI("#{@base_url}/workflows/#{opts.workflow_id}/predict_async")
+              else
+                URI("#{@url_root}/predict_async")
+              end
 
         params = {} # : Hash[Symbol | String, untyped]
-        params[:cropper] = 'true' if cropper
-        params[:full_text_ocr] = 'true' if full_text
+        params[:cropper] = 'true' if opts.cropper
+        params[:full_text_ocr] = 'true' if opts.full_text
+        params[:rag] = 'true' if opts.rag
         uri.query = URI.encode_www_form(params)
 
         headers = {
@@ -165,24 +156,23 @@ def document_queue_req_get(input_source, all_words, full_text, close_file, cropp
         }
         req = Net::HTTP::Post.new(uri, headers)
         form_data = if input_source.is_a?(Mindee::Input::Source::URLInputSource)
-                      [['document', input_source.url]]
+                      [['document', input_source.url]] # : Array[untyped]
                     else
-                      [input_source.read_contents(close: close_file)]
+                      [input_source.read_contents(close: opts.close_file)] # : Array[untyped]
                     end
-        form_data.push ['include_mvision', 'true'] if all_words
+        form_data.push ['include_mvision', 'true'] if opts.all_words
 
         req.set_form(form_data, 'multipart/form-data')
 
-        response = nil
         Net::HTTP.start(uri.hostname, uri.port, use_ssl: true, read_timeout: @request_timeout) do |http|
-          response = http.request(req)
+          return http.request(req)
         end
-        response
+        raise Mindee::Errors::MindeeError, 'Could not resolve server response.'
       end
 
       # @param job_id [String]
       # @return [Net::HTTPResponse, nil]
-      def document_queue_req(job_id)
+      def document_queue_req_get(job_id)
         uri = URI("#{@url_root}/documents/queue/#{job_id}")
 
         headers = {

lib/mindee/http/response_validation.rb

@@ -42,7 +42,7 @@ def self.valid_async_response?(response)
       # Checks and correct the response object depending on the possible kinds of returns.
       # @param response [Net::HTTPResponse]
       def self.clean_request!(response)
-        return response if (response.code.to_i < 200) || (response.code.to_i > 302)
+        return response if (response.code.to_i < 200) || (response.code.to_i > 302) # : Net::HTTPResponse
 
         return response if response.body.empty?
 

lib/mindee/parsing/common/api_response.rb

@@ -35,7 +35,7 @@ class ApiResponse
 
         # @param product_class [Mindee::Inference]
         # @param http_response [Hash]
-        # @param raw_http [String]
+        # @param raw_http [Hash]
         def initialize(product_class, http_response, raw_http)
           logger.debug('Handling API response')
           @raw_http = raw_http.to_s

lib/mindee/parsing/common/document.rb

@@ -69,7 +69,7 @@ def inject_full_text_ocr(raw_prediction)
 
           full_text_ocr = String.new
           raw_prediction.dig('inference', 'pages').each do |page|
-            full_text_ocr << (page['extras']['full_text_ocr']['content'])
+            full_text_ocr << page['extras']['full_text_ocr']['content']
           end
           artificial_text_obj = { 'content' => full_text_ocr }
           if @extras.nil? || @extras.empty?

lib/mindee/parsing/common/extras.rb

@@ -3,3 +3,4 @@
 require_relative 'extras/extras'
 require_relative 'extras/cropper_extra'
 require_relative 'extras/full_text_ocr_extra'
+require_relative 'extras/rag_extra'