WIP AI conversation limits, very quick and dirty implementation to get a prototype

app/controllers/oral_history_ai_conversation_controller.rb

@@ -16,7 +16,9 @@ def new
   end
 
   def create
-    conversation = OralHistoryAiConversationJob.launch(session_id: session.id, question: params.require(:q))
+    search_params = params.slice(:collection_limit).to_unsafe_h
+
+    conversation = OralHistoryAiConversationJob.launch(session_id: session.id, question: params.require(:q), search_params: search_params)
 
     redirect_to oral_history_ai_conversation_path(conversation.external_id)
   end

app/jobs/oral_history_ai_conversation_job.rb

@@ -1,7 +1,7 @@
 # Try to be idempotent!
 class OralHistoryAiConversationJob < ApplicationJob
-  def self.launch(question:, session_id:)
-    conversation = OralHistory::AiConversation.create!(question: question.strip, session_id: session_id)
+  def self.launch(question:, session_id:, search_params: {})
+    conversation = OralHistory::AiConversation.create!(question: question.strip, session_id: session_id, search_params: search_params)
     self.perform_later(conversation)
 
     return conversation

app/models/oral_history/ai_conversation.rb

@@ -42,7 +42,7 @@ def exec_and_record_interaction
     self.save!
 
     # Start the conversation, could take 10-20 seconds even.
-    interactor = OralHistory::ClaudeInteractor.new(question: self.question, question_embedding: self.question_embedding)
+    interactor = OralHistory::ClaudeInteractor.new(question: self.question, question_embedding: self.question_embedding, limited_start_relation: limited_start_relation)
     response = interactor.get_response(conversation_record: self)
 
     self.answer_json = interactor.extract_answer(response)
@@ -95,4 +95,24 @@ def record_chunks_used(chunks)
       }
     end
   end
+
+  # based on search_params or logged in user, we might want to limit to only certain records
+  # For now just search_params
+  #
+  # TODO WARNING this needs to be checked, there are those weird ones with requet mode "OFF" cause they
+  # aren't requestable at all now!!!
+  def limited_start_relation
+    case self.search_params["collection_limit"]
+    when "ohms"
+      OralHistoryChunk.joins(:oral_history_content).where.not(oral_history_content: { ohms_xml_text: [nil, ""]})
+    when "immediate"
+      OralHistoryChunk.joins(:oral_history_content).where(oral_history_content: { available_by_request_mode: ["off"] })
+    when "upon_request"
+      OralHistoryChunk.joins(:oral_history_content).where(oral_history_content: { available_by_request_mode: ["off", "automatic"] })
+    else
+      # unrestricted, all of em
+      OralHistoryChunk
+    end
+  end
+
 end

app/services/oral_history/chunk_fetcher.rb

@@ -8,6 +8,8 @@ class ChunkFetcher
     attr_reader :top_k, :question_embedding, :max_per_interview, :oversample_factor
     attr_reader :exclude_oral_history_chunk_ids, :exclude_oral_history_content_ids
 
+    attr_reader :limited_start_relation
+
 
     # @param top_k [Integer] how many chunks do you want back
     #
@@ -22,12 +24,14 @@ class ChunkFetcher
     #
     # @param exclude_interviews [Array<Work,OralHistoryContent,Integer>] Interviews to exclude. can be as Work, OralHistoryContent,
     #   or OralHistoryContent#id
-    def initialize(top_k:, question_embedding:, max_per_interview: nil, oversample_factor: 3, exclude_chunks: nil, exclude_interviews: nil)
+    def initialize(top_k:, question_embedding:, max_per_interview: nil, oversample_factor: 3, exclude_chunks: nil, exclude_interviews: nil, limited_start_relation:)
       @top_k = top_k
       @question_embedding = question_embedding
       @max_per_interview = max_per_interview
       @oversample_factor = oversample_factor
 
+      @limited_start_relation = limited_start_relation
+
       if exclude_chunks
         @exclude_oral_history_chunk_ids = exclude_chunks.collect {|i| i.kind_of?(OralHistoryChunk) ? i.id : i }
       end
@@ -61,7 +65,7 @@ def fetch_chunks
     # Without limit count, we'll add that later.
     def base_relation
       # Preload work, so we can get title or other metadata we might want.
-      relation = OralHistoryChunk.neighbors_for_embedding(question_embedding).includes(oral_history_content: :work)
+      relation = limited_start_relation.neighbors_for_embedding(question_embedding).includes(oral_history_content: :work)
 
       # exclude specific chunks?
       if exclude_oral_history_chunk_ids.present?

app/services/oral_history/claude_interactor.rb

@@ -20,11 +20,12 @@ class ClaudeInteractor
       region:             ScihistDigicoll::Env.lookup(:aws_region)
     )
 
-    attr_reader :question, :question_embedding
+    attr_reader :question, :question_embedding, :limited_start_relation
 
-    def initialize(question:, question_embedding:)
+    def initialize(question:, question_embedding:, limited_start_relation: OralHistoryChunk)
       @question = question
       @question_embedding = question_embedding
+      @limited_start_relation = limited_start_relation
     end
 
     # convenience to look up the embedding
@@ -107,13 +108,14 @@ def render_user_prompt(chunks)
 
     def get_chunks
       # fetch first 8 closest-vector chunks
-      chunks = OralHistory::ChunkFetcher.new(question_embedding: question_embedding, top_k: 8).fetch_chunks
+      chunks = OralHistory::ChunkFetcher.new(question_embedding: question_embedding, top_k: 8, limited_start_relation: limited_start_relation).fetch_chunks
 
       # now fetch another 8, but only 1-per-interview, not including any interviews from above
       chunks += OralHistory::ChunkFetcher.new(question_embedding: question_embedding,
                                               top_k: 8,
                                               max_per_interview: 1,
-                                              exclude_interviews: chunks.collect(&:oral_history_content_id).uniq).fetch_chunks
+                                              exclude_interviews: chunks.collect(&:oral_history_content_id).uniq,
+                                              limited_start_relation: limited_start_relation).fetch_chunks
 
       chunks
     end

app/views/oral_history_ai_conversation/new.html.erb

@@ -7,5 +7,20 @@
     <div class="d-flex justify-content-end">
       <%= f.button :submit, "Ask", class: "btn btn-brand-main" %>
     </div>
+
+    <div>
+      <%= f.input :collection_limit, as: :radio_buttons,
+            collection: [
+              ['ohms', '[OHMS-only] Only those with synchronized transcripts'],
+              ['immediate', '[All open(?)] All open for access with no verification'],
+              ['upon_request', '[semi-restricted(?) and up] Also include some that require a form to verify email address'],
+              ['permission_required', "[restricted(?) and up] Also include those that requires permission to access, after filling out a form explaining use"]
+            ],
+            label_method: ->(item) { item[1].html_safe },
+            value_method: ->(item) { item[0] },
+            checked: 'permission_required',
+            label: "Which Oral Histories to consider in search" %>
+    </div>
+
   <% end %>
 </div>

app/views/oral_history_ai_conversation/show.html.erb

@@ -19,6 +19,10 @@
     <span class="fw-bold">Q:</span> <%=  @conversation.question %>
   </div>
 
+  <div class="alert alert-info mb-4 mt-4">
+    <%= @conversation.search_params.inspect %>
+  </div>
+
   <% if [email protected]? %>
     <div class="d-flex align-items-center">
       <i class="fa fa-circle-o-notch fa-spin fa-2x me-2" aria-hidden="true"></i>

db/migrate/20251222213643_add_search_params_to_oh_ai_conversation.rb

@@ -0,0 +1,5 @@
+class AddSearchParamsToOhAiConversation < ActiveRecord::Migration[8.0]
+  def change
+    add_column :oral_history_ai_conversations, :search_params, :jsonb
+  end
+end

db/schema.rb

@@ -10,7 +10,7 @@
 #
 # It's strongly recommended that you check this file into your version control system.
 
-ActiveRecord::Schema[8.0].define(version: 2025_12_15_162911) do
+ActiveRecord::Schema[8.0].define(version: 2025_12_22_213643) do
   # These are extensions that must be enabled in order to support this database
   enable_extension "pg_catalog.plpgsql"
   enable_extension "vector"
@@ -245,6 +245,7 @@
     t.datetime "created_at", null: false
     t.datetime "updated_at", null: false
     t.string "project_source_version"
+    t.jsonb "search_params"
   end
 
   create_table "oral_history_chunks", force: :cascade do |t|