DEV: Move AI translation feature into an AI Feature (part 1)

Author: nattsw

config/locales/client.en.yml

@@ -211,6 +211,11 @@ en:
           custom_prompt: "Custom prompt"
           image_caption: "Caption images"
 
+        translation:
+          name: "Translation"
+          description: "Translates content into supported languages"
+          locale_detection: "Locale detection"
+
 
       modals:
         select_option: "Select an option..."

config/locales/server.en.yml

@@ -372,6 +372,9 @@ en:
         image_captioner:
           name: "Image captions"
           description: "Default persona powering the Helper's image caption feature"
+        locale_detection:
+          name: "Locale detection"
+          description: "Powers the translation feature by detecting the locale of a given text"
 
       topic_not_found: "Summary unavailable, topic not found!"
       summarizing: "Summarizing topic"

config/settings.yml

@@ -451,28 +451,39 @@ discourse_ai:
     default: false
     client: true
     validator: "DiscourseAi::Configuration::LlmDependencyValidator"
+    area: "ai-features/translation"
   ai_translation_model:
     default: ""
     type: enum
     allow_any: false
     enum: "DiscourseAi::Configuration::LlmEnumerator"
     validator: "DiscourseAi::Configuration::LlmValidator"
+    area: "ai-features/translation"
   ai_translation_backfill_rate:
     default: 0
+    min: 0
+    max: 1000
     client: false
     hidden: true
+    area: "ai-features/translation"
   ai_translation_backfill_limit_to_public_content:
     default: true
     client: false
-    hidden: true
+    area: "ai-features/translation"
   ai_translation_backfill_max_age_days:
     default: 5
     client: false
-    hidden: true
+    area: "ai-features/translation"
   ai_translation_verbose_logs:
     default: false
     client: false
     hidden: true
+    area: "ai-features/translation"
+  ai_translation_locale_detection_persona:
+    default: "-27"
+    type: enum
+    enum: "DiscourseAi::Configuration::PersonaEnumerator"
+    area: "ai-features/translation"
 
   inferred_concepts_enabled:
     default: false

lib/configuration/feature.rb

@@ -118,13 +118,25 @@ def ai_helper_features
           ]
         end
 
+        def translation_features
+          feature_cache[:translation] ||= [
+            new(
+              "locale_detection",
+              "ai_translation_locale_detection_persona",
+              DiscourseAi::Configuration::Module::TRANSLATION_ID,
+              DiscourseAi::Configuration::Module::TRANSLATION,
+            ),
+          ]
+        end
+
         def all
           [
             summarization_features,
             search_features,
             discord_features,
             inference_features,
             ai_helper_features,
+            translation_features,
           ].flatten
         end
 

lib/configuration/module.rb

@@ -8,14 +8,16 @@ class Module
       DISCORD = "discord"
       INFERENCE = "inference"
       AI_HELPER = "ai_helper"
+      TRANSLATION = "translation"
 
-      NAMES = [SUMMARIZATION, SEARCH, DISCORD, INFERENCE, AI_HELPER]
+      NAMES = [SUMMARIZATION, SEARCH, DISCORD, INFERENCE, AI_HELPER, TRANSLATION]
 
       SUMMARIZATION_ID = 1
       SEARCH_ID = 2
       DISCORD_ID = 3
       INFERENCE_ID = 4
       AI_HELPER_ID = 5
+      TRANSLATION_ID = 6
 
       class << self
         def all
@@ -50,6 +52,12 @@ def all
               "ai_helper_enabled",
               features: DiscourseAi::Configuration::Feature.ai_helper_features,
             ),
+            new(
+              TRANSLATION_ID,
+              TRANSLATION,
+              "ai_translation_enabled",
+              features: DiscourseAi::Configuration::Feature.translation_features,
+            ),
           ]
         end
 

lib/personas/locale_detection.rb

@@ -0,0 +1,73 @@
+# frozen_string_literal: true
+
+module DiscourseAi
+  module Personas
+    class LocaleDetection < Persona
+      def self.default_enabled
+        false
+      end
+
+      def system_prompt
+        <<~PROMPT.strip
+          You will be given a piece of text, and your task is to detect the locale (language) of the text and return it in a specific JSON format.
+
+          To complete this task, follow these steps:
+
+          1. Carefully read and analyze the provided text.
+          2. Determine the language of the text based on its characteristics, such as vocabulary, grammar, and sentence structure.
+          3. Do not use links or programing code in the text to detect the locale
+          4. Identify the appropriate language code for the detected language.
+
+          Here is a list of common language codes for reference:
+          - English: en
+          - Spanish: es
+          - French: fr
+          - German: de
+          - Italian: it
+          - Brazilian Portuguese: pt-BR
+          - Russian: ru
+          - Simplified Chinese: zh-CN
+          - Japanese: ja
+          - Korean: ko
+
+          If the language is not in this list, use the appropriate IETF language tag code.
+
+          5. Format your response as a JSON object with a single key "locale" and the value as the language code.
+
+          Your output should be in the following format:
+          <output>
+          {"locale": "xx"}
+          </output>
+
+          Where "xx" is replaced by the appropriate language code.
+
+          Important: Base your analysis solely on the provided text. Do not use any external information or make assumptions about the text's origin or context beyond what is explicitly provided.
+        PROMPT
+      end
+
+      def response_format
+        {
+          type: "json_schema",
+          json_schema: {
+            name: "reply",
+            schema: {
+              type: "object",
+              properties: {
+                locale: {
+                  type: "string",
+                },
+              },
+              required: ["locale"],
+              additionalProperties: false,
+            },
+            strict: true,
+          },
+        }
+      end
+
+      def temperature
+        0
+      end
+    end
+  end
+end

lib/personas/persona.rb

@@ -64,6 +64,7 @@ def system_personas
             Tutor => -24,
             Translator => -25,
             ImageCaptioner => -26,
+            LocaleDetection => -27,
           }
         end
 

lib/translation/language_detector.rb

@@ -4,59 +4,36 @@ module DiscourseAi
   module Translation
     class LanguageDetector
       DETECTION_CHAR_LIMIT = 1000
-      PROMPT_TEXT = <<~TEXT
-      You will be given a piece of text, and your task is to detect the locale (language) of the text and return it in a specific JSON format.
-
-      To complete this task, follow these steps:
-
-      1. Carefully read and analyze the provided text.
-      2. Determine the language of the text based on its characteristics, such as vocabulary, grammar, and sentence structure.
-      3. Do not use links or programing code in the text to detect the locale
-      4. Identify the appropriate language code for the detected language.
-
-      Here is a list of common language codes for reference:
-      - English: en
-      - Spanish: es
-      - French: fr
-      - German: de
-      - Italian: it
-      - Brazilian Portuguese: pt-BR
-      - Russian: ru
-      - Simplified Chinese: zh-CN
-      - Japanese: ja
-      - Korean: ko
-
-      If the language is not in this list, use the appropriate IETF language tag code.
-
-      5. Format your response as a JSON object with a single key "locale" and the value as the language code.
-
-      Your output should be in the following format:
-      <output>
-      {"locale": "xx"}
-      </output>
-
-      Where "xx" is replaced by the appropriate language code.
-
-      Important: Base your analysis solely on the provided text. Do not use any external information or make assumptions about the text's origin or context beyond what is explicitly provided.
-    TEXT
 
       def initialize(text)
         @text = text
       end
 
       def detect
+        return nil if !SiteSetting.ai_translation_enabled
+        if (
+             ai_persona = AiPersona.find_by(id: SiteSetting.ai_translation_locale_detection_persona)
+           ).blank?
+          return nil
+        end
+
+        persona_klass = ai_persona.class_instance
+
+        llm_model = preferred_llm_model(ai_persona, persona_klass)
+        return nil if llm_model.blank?
+
         prompt =
           DiscourseAi::Completions::Prompt.new(
-            PROMPT_TEXT,
+            ai_persona.system_prompt,
             messages: [{ type: :user, content: @text, id: "user" }],
           )
-
+        response_format = persona_klass.new.response_format
         structured_output =
-          DiscourseAi::Completions::Llm.proxy(SiteSetting.ai_translation_model).generate(
+          DiscourseAi::Completions::Llm.proxy(llm_model).generate(
             prompt,
-            user: Discourse.system_user,
+            user: ai_persona.user || Discourse.system_user,
             feature_name: "translation",
-            response_format: response_format,
+            response_format:,
           )
 
         structured_output&.read_buffered_property(:locale)
@@ -81,6 +58,16 @@ def response_format
           },
         }
       end
+
+      private
+
+      def preferred_llm_model(ai_persona, persona_klass)
+        if ai_persona.force_default_llm
+          persona_klass.default_llm_id
+        else
+          SiteSetting.ai_translation_model.presence || persona_klass.default_llm_id
+        end
+      end
     end
   end
 end

spec/lib/translation/language_detector_spec.rb

@@ -13,7 +13,7 @@
 
     it "creates the correct prompt" do
       allow(DiscourseAi::Completions::Prompt).to receive(:new).with(
-        DiscourseAi::Translation::LanguageDetector::PROMPT_TEXT,
+        AiPersona.find_by(id: SiteSetting.ai_translation_locale_detection_persona).system_prompt,
         messages: [{ type: :user, content: "meow", id: "user" }],
       ).and_call_original
 

spec/requests/admin/ai_features_controller_spec.rb

@@ -19,7 +19,7 @@
       get "/admin/plugins/discourse-ai/ai-features.json"
 
       expect(response.status).to eq(200)
-      expect(response.parsed_body["ai_features"].count).to eq(5)
+      expect(response.parsed_body["ai_features"].count).to eq(6)
     end
   end