ragdoll 0.1.3 → 0.1.9

data/app/models/ragdoll/search.rb

@@ -0,0 +1,165 @@
+# frozen_string_literal: true
+
+require "active_record"
+require "neighbor"
+
+module Ragdoll
+  class Search < ActiveRecord::Base
+    self.table_name = "ragdoll_searches"
+
+    # Use pgvector for vector similarity search on query embeddings
+    has_neighbors :query_embedding
+
+    has_many :search_results, class_name: "Ragdoll::SearchResult", foreign_key: "search_id", dependent: :destroy
+    has_many :embeddings, through: :search_results
+
+    validates :query, presence: true
+    validates :query_embedding, presence: true
+    validates :search_type, presence: true, inclusion: { in: %w[semantic hybrid fulltext] }
+    validates :results_count, presence: true, numericality: { greater_than_or_equal_to: 0 }
+
+    scope :by_type, ->(type) { where(search_type: type) }
+    scope :by_session, ->(session_id) { where(session_id: session_id) }
+    scope :by_user, ->(user_id) { where(user_id: user_id) }
+    scope :recent, -> { order(created_at: :desc) }
+    scope :with_results, -> { where("results_count > 0") }
+    scope :popular, -> { where("results_count > 0").order(results_count: :desc) }
+    scope :slow_searches, ->(threshold_ms = 1000) { where("execution_time_ms > ?", threshold_ms) }
+
+    # Find searches with similar query embeddings
+    def self.find_similar(query_embedding, limit: 10, threshold: 0.8)
+      nearest_neighbors(:query_embedding, query_embedding, distance: "cosine")
+        .limit(limit * 2)
+        .map do |search|
+          similarity = 1.0 - search.neighbor_distance
+          next if similarity < threshold
+          
+          search.define_singleton_method(:similarity_score) { similarity }
+          search
+        end
+        .compact
+        .sort_by(&:similarity_score)
+        .reverse
+        .take(limit)
+    end
+
+    # Calculate statistics for this search
+    def calculate_similarity_stats!
+      return unless search_results.any?
+      
+      scores = search_results.pluck(:similarity_score)
+      update!(
+        max_similarity_score: scores.max,
+        min_similarity_score: scores.min,
+        avg_similarity_score: scores.sum.to_f / scores.length
+      )
+    end
+
+    # Get search results ordered by rank
+    def ranked_results
+      search_results.includes(:embedding).order(:result_rank)
+    end
+
+    # Get clicked results
+    def clicked_results
+      search_results.where(clicked: true).order(:clicked_at)
+    end
+
+    # Calculate click-through rate
+    def click_through_rate
+      return 0.0 if results_count == 0
+      
+      clicked_count = search_results.where(clicked: true).count
+      clicked_count.to_f / results_count
+    end
+
+    # Record a search with its results
+    def self.record_search(query:, query_embedding:, results:, search_type: "semantic", 
+                          filters: {}, options: {}, execution_time_ms: nil, 
+                          session_id: nil, user_id: nil)
+      search = create!(
+        query: query,
+        query_embedding: query_embedding,
+        search_type: search_type,
+        results_count: results.length,
+        search_filters: filters,
+        search_options: options,
+        execution_time_ms: execution_time_ms,
+        session_id: session_id,
+        user_id: user_id
+      )
+
+      # Create search result records
+      results.each_with_index do |result, index|
+        search.search_results.create!(
+          embedding_id: result[:embedding_id],
+          similarity_score: result[:similarity],
+          result_rank: index + 1
+        )
+      end
+
+      # Calculate and store similarity statistics
+      search.calculate_similarity_stats!
+      search
+    end
+
+    # Search analytics methods
+    def self.search_analytics(days: 30)
+      start_date = days.days.ago
+      searches = where(created_at: start_date..)
+      
+      {
+        total_searches: searches.count,
+        unique_queries: searches.distinct.count(:query),
+        avg_results_per_search: searches.average(:results_count)&.round(2),
+        avg_execution_time: searches.average(:execution_time_ms)&.round(2),
+        search_types: searches.group(:search_type).count,
+        searches_with_results: searches.where("results_count > 0").count,
+        avg_click_through_rate: calculate_avg_ctr(searches)
+      }
+    end
+
+    # Cleanup orphaned searches that have no remaining search results
+    def self.cleanup_orphaned_searches
+      orphaned_search_ids = where.not(id: SearchResult.distinct.pluck(:search_id))
+      orphaned_count = orphaned_search_ids.count
+      
+      if orphaned_count > 0
+        orphaned_search_ids.destroy_all
+        Rails.logger.info "Cleaned up #{orphaned_count} orphaned search records" if defined?(Rails)
+      end
+      
+      orphaned_count
+    end
+
+    # Cleanup searches older than specified days with no clicks
+    def self.cleanup_old_unused_searches(days: 30)
+      cutoff_date = days.days.ago
+      unused_searches = where(created_at: ...cutoff_date)
+                       .left_joins(:search_results)
+                       .where(search_results: { clicked: [nil, false] })
+      
+      unused_count = unused_searches.count
+      
+      if unused_count > 0
+        unused_searches.destroy_all
+        Rails.logger.info "Cleaned up #{unused_count} old unused search records" if defined?(Rails)
+      end
+      
+      unused_count
+    end
+
+    private
+
+    def self.calculate_avg_ctr(searches)
+      search_ids = searches.pluck(:id)
+      return 0.0 if search_ids.empty?
+
+      total_results = SearchResult.where(search_id: search_ids).count
+      return 0.0 if total_results == 0
+
+      clicked_results = SearchResult.where(search_id: search_ids, clicked: true).count
+      (clicked_results.to_f / total_results * 100).round(2)
+    end
+  end
+end

data/app/models/ragdoll/search_result.rb

@@ -0,0 +1,121 @@
+# frozen_string_literal: true
+
+require "active_record"
+
+module Ragdoll
+  class SearchResult < ActiveRecord::Base
+    self.table_name = "ragdoll_search_results"
+
+    belongs_to :search, class_name: "Ragdoll::Search"
+    belongs_to :embedding, class_name: "Ragdoll::Embedding"
+
+    validates :similarity_score, presence: true, numericality: { in: 0.0..1.0 }
+    validates :result_rank, presence: true, numericality: { greater_than: 0 }
+    validates :result_rank, uniqueness: { scope: :search_id }
+
+    scope :by_rank, -> { order(:result_rank) }
+    scope :clicked, -> { where(clicked: true) }
+    scope :unclicked, -> { where(clicked: false) }
+    scope :high_similarity, ->(threshold = 0.8) { where("similarity_score >= ?", threshold) }
+    scope :recent_clicks, -> { where(clicked: true).order(clicked_at: :desc) }
+
+    # Cleanup callback to remove searches when they have no results left
+    after_destroy :cleanup_empty_search
+
+    # Mark this result as clicked
+    def mark_as_clicked!
+      update!(clicked: true, clicked_at: Time.current)
+    end
+
+    # Get the content through the embedding relationship
+    def content
+      embedding&.content
+    end
+
+    # Get the document through the embedding relationship
+    def document
+      embedding&.embeddable&.document
+    end
+
+    # Get the document title
+    def document_title
+      document&.title
+    end
+
+    # Get the document location
+    def document_location
+      document&.location
+    end
+
+    # Analytics for search results
+    def self.analytics(days: 30)
+      start_date = days.days.ago
+      results = where(created_at: start_date..)
+      
+      {
+        total_results: results.count,
+        clicked_results: results.where(clicked: true).count,
+        click_through_rate: calculate_ctr(results),
+        avg_similarity_score: results.average(:similarity_score)&.round(4),
+        high_similarity_results: results.where("similarity_score >= 0.8").count,
+        low_similarity_results: results.where("similarity_score < 0.5").count,
+        rank_performance: rank_click_analysis(results)
+      }
+    end
+
+    # Analyze click performance by result rank
+    def self.rank_click_analysis(results = nil)
+      results ||= all
+      
+      results.group(:result_rank)
+             .group("clicked")
+             .count
+             .each_with_object({}) do |((rank, clicked), count), hash|
+        hash[rank] ||= { total: 0, clicked: 0 }
+        hash[rank][:total] += count
+        hash[rank][:clicked] += count if clicked
+      end
+             .transform_values do |stats|
+        stats.merge(
+          ctr: stats[:total] > 0 ? (stats[:clicked].to_f / stats[:total] * 100).round(2) : 0.0
+        )
+      end
+    end
+
+    # Find embeddings that perform well across multiple searches
+    def self.top_performing_embeddings(limit: 20)
+      joins(:embedding)
+        .group(:embedding_id)
+        .select(
+          "embedding_id",
+          "COUNT(*) as appearance_count",
+          "AVG(similarity_score) as avg_similarity",
+          "COUNT(CASE WHEN clicked THEN 1 END) as click_count",
+          "ROUND(COUNT(CASE WHEN clicked THEN 1 END) * 100.0 / COUNT(*), 2) as ctr"
+        )
+        .having("COUNT(*) > 1")
+        .order("avg_similarity DESC, ctr DESC")
+        .limit(limit)
+    end
+
+    private
+
+    def self.calculate_ctr(results)
+      total = results.count
+      return 0.0 if total == 0
+
+      clicked = results.where(clicked: true).count
+      (clicked.to_f / total * 100).round(2)
+    end
+
+    # Cleanup callback to remove parent search if it has no results left
+    def cleanup_empty_search
+      return unless search
+      
+      # Check if this was the last result for the search
+      if search.search_results.count == 0
+        search.destroy
+      end
+    end
+  end
+end

data/app/services/ragdoll/configuration_service.rb

@@ -20,10 +20,10 @@ module Ragdoll
         @config.embedding_model(content_type)
       when :summary, :keywords
         # Check for task-specific model, fall back to default
-        task_model = @config.models.text_generation[task_type]
-        task_model || @config.models.text_generation[:default]
+        task_model = @config.models[:text_generation][task_type]
+        task_model || @config.models[:text_generation][:default]
       else
-        @config.models.text_generation[:default]
+        @config.models[:text_generation][:default]
       end
     end
 

data/app/services/ragdoll/document_processor.rb

@@ -3,6 +3,8 @@
 require "pdf-reader"
 require "docx"
 require "rmagick"
+require "yaml"
+require "date"
 # Image description service is auto-loaded from app/services
 
 module Ragdoll
@@ -137,6 +139,11 @@ module Ragdoll
         raise ParseError, "Unsupported PDF feature: #{e.message}"
       end
 
+      # Add filepath-based title as fallback if no title was found
+      if metadata[:title].nil? || (metadata[:title].is_a?(String) && metadata[:title].strip.empty?)
+        metadata[:title] = extract_title_from_filepath
+      end
+
       {
         content: content.strip,
         metadata: metadata,
@@ -192,6 +199,11 @@ module Ragdoll
         raise ParseError, "#{__LINE__} Failed to parse DOCX: #{e.message}"
       end
 
+      # Add filepath-based title as fallback if no title was found
+      if metadata[:title].nil? || (metadata[:title].is_a?(String) && metadata[:title].strip.empty?)
+        metadata[:title] = extract_title_from_filepath
+      end
+
       {
         content: content.strip,
         metadata: metadata,
@@ -212,6 +224,20 @@ module Ragdoll
                       else "text"
                       end
 
+      # Parse YAML front matter for markdown files
+      if document_type == "markdown" && content.start_with?("---\n")
+        front_matter, body_content = parse_yaml_front_matter(content)
+        if front_matter
+          metadata.merge!(front_matter)
+          content = body_content
+        end
+      end
+
+      # Add filepath-based title as fallback if no title was found
+      if metadata[:title].nil? || (metadata[:title].is_a?(String) && metadata[:title].strip.empty?)
+        metadata[:title] = extract_title_from_filepath
+      end
+
       {
         content: content,
         metadata: metadata,
@@ -225,16 +251,41 @@ module Ragdoll
         encoding: "ISO-8859-1"
       }
 
+      # Try to parse front matter with different encoding too
+      if document_type == "markdown" && content.start_with?("---\n")
+        front_matter, body_content = parse_yaml_front_matter(content)
+        if front_matter
+          metadata.merge!(front_matter)
+          content = body_content
+        end
+      end
+
+      # Add filepath-based title as fallback if no title was found
+      if metadata[:title].nil? || (metadata[:title].is_a?(String) && metadata[:title].strip.empty?)
+        metadata[:title] = extract_title_from_filepath
+      end
+
       {
         content: content,
         metadata: metadata,
-        document_type: "text"
+        document_type: document_type.nil? ? "text" : document_type
       }
     end
 
     def parse_html
       content = File.read(@file_path, encoding: "UTF-8")
 
+      # Extract title from H1 tag if present
+      h1_match = content.match(%r{<h1[^>]*>(.*?)</h1>}mi)
+      title = nil
+      if h1_match
+        # Clean up the H1 content by removing any HTML tags and normalizing whitespace
+        title = h1_match[1]
+                  .gsub(/<[^>]+>/, " ")  # Remove any nested HTML tags
+                  .gsub(/\s+/, " ")      # Normalize whitespace
+                  .strip
+      end
+
       # Basic HTML tag stripping (for more advanced parsing, consider using Nokogiri)
       clean_content = content
                       .gsub(%r{<script[^>]*>.*?</script>}mi, "") # Remove script tags
@@ -248,6 +299,13 @@ module Ragdoll
         original_format: "html"
       }
 
+      # Add title to metadata if found, otherwise use filepath fallback
+      if title && !title.empty?
+        metadata[:title] = title
+      else
+        metadata[:title] = extract_title_from_filepath
+      end
+
       {
         content: clean_content,
         metadata: metadata,
@@ -286,6 +344,9 @@ module Ragdoll
       # Use AI-generated description or fallback placeholder
       content = desc && !desc.empty? ? desc : "Image file: #{File.basename(@file_path)}"
 
+      # Add filepath-based title as fallback
+      metadata[:title] = extract_title_from_filepath
+
       puts "✅ DocumentProcessor: Image parsing complete. Content: '#{content[0..100]}...'"
 
       {
@@ -338,5 +399,67 @@ module Ragdoll
       else "application/octet-stream"
       end
     end
+
+    private
+
+    # Extract a meaningful title from the file path as a fallback
+    # @param file_path [String] the full file path
+    # @return [String] a cleaned title derived from the filename
+    def extract_title_from_filepath(file_path = @file_path)
+      filename = File.basename(file_path, File.extname(file_path))
+      
+      # Clean up common patterns in filenames to make them more readable
+      title = filename
+               .gsub(/[-_]+/, ' ')           # Replace hyphens and underscores with spaces
+               .gsub(/([a-z])([A-Z])/, '\1 \2') # Add space before capital letters (camelCase)
+               .gsub(/\s+/, ' ')             # Normalize multiple spaces
+               .strip
+      
+      # Capitalize words for better readability
+      title.split(' ').map(&:capitalize).join(' ')
+    end
+
+    # Parse YAML front matter from markdown content
+    # @param content [String] the full content of the markdown file
+    # @return [Array] returns [front_matter_hash, body_content] or [nil, original_content]
+    def parse_yaml_front_matter(content)
+      # Check if content starts with YAML front matter delimiter
+      return [nil, content] unless content.start_with?("---\n")
+
+      # Find the closing delimiter
+      lines = content.lines
+      closing_index = nil
+      
+      lines.each_with_index do |line, index|
+        next if index == 0 # Skip the opening ---
+        if line.strip == "---"
+          closing_index = index
+          break
+        end
+      end
+
+      # No closing delimiter found
+      return [nil, content] unless closing_index
+
+      # Extract YAML content and body
+      yaml_lines = lines[1...closing_index]
+      body_lines = lines[(closing_index + 1)..-1]
+
+      yaml_content = yaml_lines.join
+      body_content = body_lines&.join || ""
+
+      # Parse YAML
+      begin
+        # Allow Time objects for date fields in YAML front matter
+        front_matter = YAML.safe_load(yaml_content, permitted_classes: [Time, Date])
+        # Convert string keys to symbols for consistency
+        front_matter = front_matter.transform_keys(&:to_sym) if front_matter.is_a?(Hash)
+        [front_matter, body_content.strip]
+      rescue YAML::SyntaxError, Psych::DisallowedClass => e
+        # If YAML parsing fails, return original content
+        Rails.logger.warn "Warning: Failed to parse YAML front matter: #{e.message}" if defined?(Rails)
+        [nil, content]
+      end
+    end
   end
 end

data/app/services/ragdoll/embedding_service.rb

@@ -38,6 +38,11 @@ module Ragdoll
           embedding_config = @model_resolver.resolve_embedding(:text)
           # Use just the model name for RubyLLM
           model = embedding_config.model.model
+          
+          # If model is nil or empty, use fallback
+          if model.nil? || model.empty?
+            return generate_fallback_embedding
+          end
 
           begin
             response = RubyLLM.embed(cleaned_text, model: model)
@@ -93,6 +98,11 @@ module Ragdoll
           embedding_config = @model_resolver.resolve_embedding(:text)
           # Use just the model name for RubyLLM
           model = embedding_config.model.model
+          
+          # If model is nil or empty, use fallback
+          if model.nil? || model.empty?
+            return cleaned_texts.map { generate_fallback_embedding }
+          end
 
           cleaned_texts.map do |text|
             response = RubyLLM.embed(text, model: model)

data/app/services/ragdoll/search_engine.rb

@@ -27,25 +27,83 @@ module Ragdoll
     end
 
     def search_similar_content(query_or_embedding, options = {})
+      start_time = Time.current
       search_config = @config_service.search_config
       limit = options[:limit] || search_config[:max_results]
       threshold = options[:threshold] || search_config[:similarity_threshold]
       filters = options[:filters] || {}
+      
+      # Extract tracking options
+      session_id = options[:session_id]
+      user_id = options[:user_id]
+      track_search = options.fetch(:track_search, true)
 
       if query_or_embedding.is_a?(Array)
         # It's already an embedding
         query_embedding = query_or_embedding
+        query_string = options[:query] # Should be provided when passing embedding directly
       else
         # It's a query string, generate embedding
-        query_embedding = @embedding_service.generate_embedding(query_or_embedding)
+        query_string = query_or_embedding
+        query_embedding = @embedding_service.generate_embedding(query_string)
         return [] if query_embedding.nil?
       end
 
-      # Search using ActiveRecord models
-      Ragdoll::Embedding.search_similar(query_embedding,
-                                       limit: limit,
-                                       threshold: threshold,
-                                       filters: filters)
+      # Search using ActiveRecord models with statistics
+      # Try enhanced search first, fall back to original if it fails
+      begin
+        search_response = Ragdoll::Embedding.search_similar_with_stats(query_embedding,
+                                                                      limit: limit,
+                                                                      threshold: threshold,
+                                                                      filters: filters)
+        results = search_response[:results]
+        statistics = search_response[:statistics]
+      rescue NoMethodError, PG::SyntaxError => e
+        # Fall back to original search method if enhanced version fails
+        puts "Warning: Enhanced search failed (#{e.message}), using fallback" if ENV["RAGDOLL_DEBUG"]
+        results = Ragdoll::Embedding.search_similar(query_embedding,
+                                                   limit: limit,
+                                                   threshold: threshold,
+                                                   filters: filters)
+        statistics = nil
+      end
+      
+      execution_time = ((Time.current - start_time) * 1000).round
+      
+      # Record search if tracking enabled and we have a query string
+      if track_search && query_string && !query_string.empty?
+        begin
+          # Format results for search recording
+          search_results = results.map do |result|
+            {
+              embedding_id: result[:embedding_id] || result[:id],
+              similarity: result[:similarity] || result[:similarity_score] || 0.0
+            }
+          end
+          
+          Ragdoll::Search.record_search(
+            query: query_string,
+            query_embedding: query_embedding,
+            results: search_results,
+            search_type: "semantic",
+            filters: filters,
+            options: { limit: limit, threshold: threshold },
+            execution_time_ms: execution_time,
+            session_id: session_id,
+            user_id: user_id
+          )
+        rescue => e
+          # Log error but don't fail the search
+          puts "Warning: Search tracking failed: #{e.message}" if ENV["RAGDOLL_DEBUG"]
+        end
+      end
+
+      # Return results with statistics for better user feedback
+      {
+        results: results,
+        statistics: statistics,
+        execution_time_ms: execution_time
+      }
     end
   end
 end

data/db/migrate/007_create_ragdoll_searches.rb

@@ -0,0 +1,73 @@
+class CreateRagdollSearches < ActiveRecord::Migration[7.0]
+  def change
+    create_table :ragdoll_searches,
+      comment: "Search queries and results tracking with vector similarity support" do |t|
+
+      t.text :query, null: false,
+        comment: "Original search query text"
+
+      t.vector :query_embedding, limit: 1536, null: false,
+        comment: "Vector embedding of the search query for similarity matching"
+
+      t.string :search_type, null: false, default: "semantic",
+        comment: "Type of search performed (semantic, hybrid, fulltext)"
+
+      t.integer :results_count, null: false, default: 0,
+        comment: "Number of results returned for this search"
+
+      t.float :max_similarity_score, 
+        comment: "Highest similarity score from results"
+
+      t.float :min_similarity_score,
+        comment: "Lowest similarity score from results"
+
+      t.float :avg_similarity_score,
+        comment: "Average similarity score of results"
+
+      t.json :search_filters, default: {},
+        comment: "Filters applied during search (document_type, date_range, etc.)"
+
+      t.json :search_options, default: {},
+        comment: "Search configuration options (threshold, limit, etc.)"
+
+      t.integer :execution_time_ms,
+        comment: "Search execution time in milliseconds"
+
+      t.string :session_id,
+        comment: "User session identifier for grouping related searches"
+
+      t.string :user_id,
+        comment: "User identifier if authentication is available"
+
+      t.timestamps null: false,
+        comment: "Standard creation and update timestamps"
+
+      ###########
+      # Indexes #
+      ###########
+
+      t.index :query_embedding, using: :ivfflat, opclass: :vector_cosine_ops, 
+        name: "index_ragdoll_searches_on_query_embedding_cosine",
+        comment: "IVFFlat index for finding similar search queries"
+
+      t.index :search_type,
+        comment: "Index for filtering by search type"
+
+      t.index :session_id,
+        comment: "Index for grouping searches by session"
+
+      t.index :user_id,
+        comment: "Index for filtering searches by user"
+
+      t.index :created_at,
+        comment: "Index for chronological search history"
+
+      t.index :results_count,
+        comment: "Index for analyzing search effectiveness"
+
+      t.index "to_tsvector('english', query)", using: :gin, 
+        name: "index_ragdoll_searches_on_fulltext_query",
+        comment: "Full-text search index for finding searches by query text"
+    end
+  end
+end