diff --git a/config/settings.yml b/config/settings.yml
index 8ab7cc4c..7240fa06 100644
--- a/config/settings.yml
+++ b/config/settings.yml
@@ -1,5 +1,5 @@
 plugins:
-  ai_enabled:
+  discourse_ai_enabled:
     default: true
     client: true
 
@@ -64,3 +64,10 @@ plugins:
     choices:
      - sentiment
      - emotion
+  
+  ai_nsfw_live_detection_enabled: false
+  ai_nsfw_inference_service_api_endpoint:
+    default: "https://nsfw-testing.demo-by-discourse.com"
+  ai_nsfw_inference_service_api_key:
+    default: ""
+  ai_nsfw_probability_threshold: 60
diff --git a/lib/modules/nsfw/entry_point.rb b/lib/modules/nsfw/entry_point.rb
new file mode 100644
index 00000000..a0eeb280
--- /dev/null
+++ b/lib/modules/nsfw/entry_point.rb
@@ -0,0 +1,16 @@
+# frozen_string_literal: true
+
+module DiscourseAI
+  module NSFW
+    class EntryPoint
+      def inject_into(plugin)
+        require_relative "evaluation.rb"
+        require_relative "jobs/regular/evaluate_content.rb"
+
+        plugin.add_model_callback(Upload, :after_create) do
+          Jobs.enqueue(:evaluate_content, upload_id: self.id)
+        end
+      end
+    end
+  end
+end
diff --git a/lib/modules/nsfw/evaluation.rb b/lib/modules/nsfw/evaluation.rb
new file mode 100644
index 00000000..286734fc
--- /dev/null
+++ b/lib/modules/nsfw/evaluation.rb
@@ -0,0 +1,38 @@
+# frozen_string_literal: true
+
+module DiscourseAI
+  module NSFW
+    class Evaluation
+      AVAILABLE_MODELS = %w[opennsfw2 nsfw_detector]
+
+      def perform(upload)
+        result = { verdict: false, evaluation: {} }
+
+        AVAILABLE_MODELS.each do |model|
+          model_result = evaluate_with_model(model, upload).symbolize_keys!
+
+          model_result.values.each do |classification_prob|
+            if classification_prob.to_i >= SiteSetting.ai_nsfw_probability_threshold
+              result[:verdict] = true
+            end
+          end
+
+          result[:evaluation][model.to_sym] = model_result
+        end
+
+        result
+      end
+
+      private
+
+      def evaluate_with_model(model, upload)
+        DiscourseAI::InferenceManager.perform!(
+          "#{SiteSetting.ai_nsfw_inference_service_api_endpoint}/api/v1/classify",
+          model,
+          Discourse.store.cdn_url(upload.url),
+          SiteSetting.ai_nsfw_inference_service_api_key,
+        )
+      end
+    end
+  end
+end
diff --git a/lib/modules/nsfw/jobs/regular/evaluate_content.rb b/lib/modules/nsfw/jobs/regular/evaluate_content.rb
new file mode 100644
index 00000000..4976df83
--- /dev/null
+++ b/lib/modules/nsfw/jobs/regular/evaluate_content.rb
@@ -0,0 +1,17 @@
+# frozen_string_literal: true
+
+module Jobs
+  class EvaluateContent < ::Jobs::Base
+    def execute(args)
+      upload = Upload.find_by_id(args[:upload_id])
+
+      return unless upload
+
+      result = DiscourseAI::NSFW::Evaluation.new.perform(upload)
+
+      # FIXME(roman): This is a simplistic action just to create
+      # the basic flow. We'll introduce flagging capabilities in the future.
+      upload.destroy! if result[:verdict]
+    end
+  end
+end
diff --git a/plugin.rb b/plugin.rb
index 755f5a49..afbad8d8 100644
--- a/plugin.rb
+++ b/plugin.rb
@@ -7,10 +7,10 @@
 # url: TODO
 # required_version: 2.7.0
 
-enabled_site_setting :ai_enabled
+enabled_site_setting :discourse_ai_enabled
 
 after_initialize do
-  module ::Disorder
+  module ::DiscourseAI
     PLUGIN_NAME = "discourse-ai"
   end
 
@@ -40,4 +40,7 @@ after_initialize do
   on(:chat_message_edited) do |chat_message|
     DiscourseAI::Toxicity::EventHandler.handle_chat_async(chat_message)
   end
+
+  require_relative "lib/modules/nsfw/entry_point.rb"
+  DiscourseAI::NSFW::EntryPoint.new.inject_into(self)
 end
diff --git a/spec/.gitkeep b/spec/.gitkeep
deleted file mode 100644
index e69de29b..00000000
diff --git a/spec/lib/modules/nsfw/evaluation_spec.rb b/spec/lib/modules/nsfw/evaluation_spec.rb
new file mode 100644
index 00000000..2ca9dbd0
--- /dev/null
+++ b/spec/lib/modules/nsfw/evaluation_spec.rb
@@ -0,0 +1,46 @@
+# frozen_string_literal: true
+
+require "rails_helper"
+require_relative "../../../support/nsfw_inference_stubs"
+
+describe DiscourseAI::NSFW::Evaluation do
+  before { SiteSetting.ai_nsfw_live_detection_enabled = true }
+
+  fab!(:image) { Fabricate(:s3_image_upload) }
+
+  let(:available_models) { DiscourseAI::NSFW::Evaluation::AVAILABLE_MODELS }
+
+  describe "perform" do
+    context "when we determine content is NSFW" do
+      before { NSFWInferenceStubs.positive(image) }
+
+      it "returns true alongside the evaluation" do
+        result = subject.perform(image)
+
+        expect(result[:verdict]).to eq(true)
+
+        available_models.each do |model|
+          expect(result.dig(:evaluation, model.to_sym)).to eq(
+            NSFWInferenceStubs.positive_result(model),
+          )
+        end
+      end
+    end
+
+    context "when we determine content is safe" do
+      before { NSFWInferenceStubs.negative(image) }
+
+      it "returns false alongside the evaluation" do
+        result = subject.perform(image)
+
+        expect(result[:verdict]).to eq(false)
+
+        available_models.each do |model|
+          expect(result.dig(:evaluation, model.to_sym)).to eq(
+            NSFWInferenceStubs.negative_result(model),
+          )
+        end
+      end
+    end
+  end
+end
diff --git a/spec/lib/modules/nsfw/jobs/regular/evaluate_content_spec.rb b/spec/lib/modules/nsfw/jobs/regular/evaluate_content_spec.rb
new file mode 100644
index 00000000..c74c0a68
--- /dev/null
+++ b/spec/lib/modules/nsfw/jobs/regular/evaluate_content_spec.rb
@@ -0,0 +1,30 @@
+# frozen_string_literal: true
+
+require "rails_helper"
+require_relative "../../../../../support/nsfw_inference_stubs"
+
+describe Jobs::EvaluateContent do
+  fab!(:image) { Fabricate(:s3_image_upload) }
+
+  describe "#execute" do
+    context "when we conclude content is NSFW" do
+      before { NSFWInferenceStubs.positive(image) }
+
+      it "deletes the upload" do
+        subject.execute(upload_id: image.id)
+
+        expect { image.reload }.to raise_error(ActiveRecord::RecordNotFound)
+      end
+    end
+
+    context "when we conclude content is not NSFW" do
+      before { NSFWInferenceStubs.negative(image) }
+
+      it "does nothing" do
+        subject.execute(upload_id: image.id)
+
+        expect(image.reload).to be_present
+      end
+    end
+  end
+end
diff --git a/spec/support/nsfw_inference_stubs.rb b/spec/support/nsfw_inference_stubs.rb
new file mode 100644
index 00000000..6d43cabb
--- /dev/null
+++ b/spec/support/nsfw_inference_stubs.rb
@@ -0,0 +1,45 @@
+class NSFWInferenceStubs
+  class << self
+    def endpoint
+      "#{SiteSetting.ai_nsfw_inference_service_api_endpoint}/api/v1/classify"
+    end
+
+    def upload_url(upload)
+      Discourse.store.cdn_url(upload.url)
+    end
+
+    def positive_result(model)
+      return { nsfw_probability: 90 } if model == "opennsfw2"
+      { drawings: 1, hentai: 2, neutral: 0, porn: 90, sexy: 79 }
+    end
+
+    def negative_result(model)
+      return { nsfw_probability: 3 } if model == "opennsfw2"
+      { drawings: 1, hentai: 2, neutral: 0, porn: 3, sexy: 1 }
+    end
+
+    def positive(upload)
+      WebMock
+        .stub_request(:post, endpoint)
+        .with(body: JSON.dump(model: "nsfw_detector", content: upload_url(upload)))
+        .to_return(status: 200, body: JSON.dump(positive_result("nsfw_detector")))
+
+      WebMock
+        .stub_request(:post, endpoint)
+        .with(body: JSON.dump(model: "opennsfw2", content: upload_url(upload)))
+        .to_return(status: 200, body: JSON.dump(positive_result("opennsfw2")))
+    end
+
+    def negative(upload)
+      WebMock
+        .stub_request(:post, endpoint)
+        .with(body: JSON.dump(model: "nsfw_detector", content: upload_url(upload)))
+        .to_return(status: 200, body: JSON.dump(negative_result("nsfw_detector")))
+
+      WebMock
+        .stub_request(:post, endpoint)
+        .with(body: JSON.dump(model: "opennsfw2", content: upload_url(upload)))
+        .to_return(status: 200, body: JSON.dump(negative_result("opennsfw2")))
+    end
+  end
+end