leva 0.1.9.1 → 0.1.11

data/app/controllers/leva/application_controller.rb

@@ -4,4 +4,4 @@ module Leva
   class ApplicationController < ActionController::Base
     protect_from_forgery with: :exception
   end
-end
+end

data/app/controllers/leva/dataset_records_controller.rb

@@ -18,4 +18,4 @@ module Leva
       @dataset = Dataset.find(params[:dataset_id])
     end
   end
-end
+end

data/app/controllers/leva/datasets_controller.rb

@@ -2,7 +2,7 @@
 
 module Leva
   class DatasetsController < ApplicationController
-    before_action :set_dataset, only: [:show, :edit, :update, :destroy]
+    before_action :set_dataset, only: [ :show, :edit, :update, :destroy ]
 
     # GET /datasets
     # @return [void]
@@ -35,7 +35,7 @@ module Leva
       @dataset = Dataset.new(dataset_params)
 
       if @dataset.save
-        redirect_to @dataset, notice: 'Dataset was successfully created.'
+        redirect_to @dataset, notice: "Dataset was successfully created."
       else
         render :new
       end
@@ -45,7 +45,7 @@ module Leva
     # @return [void]
     def update
       if @dataset.update(dataset_params)
-        redirect_to @dataset, notice: 'Dataset was successfully updated.'
+        redirect_to @dataset, notice: "Dataset was successfully updated."
       else
         render :edit
       end
@@ -55,10 +55,10 @@ module Leva
     # @return [void]
     def destroy
       if @dataset.dataset_records.any?
-        redirect_to @dataset, alert: 'Cannot delete dataset with existing records.'
+        redirect_to @dataset, alert: "Cannot delete dataset with existing records."
       else
         @dataset.destroy
-        redirect_to datasets_url, notice: 'Dataset was successfully destroyed.'
+        redirect_to datasets_url, notice: "Dataset was successfully destroyed."
       end
     end
 
@@ -76,4 +76,4 @@ module Leva
       params.require(:dataset).permit(:name, :description)
     end
   end
-end
+end

data/app/controllers/leva/design_system_controller.rb

@@ -0,0 +1,9 @@
+# frozen_string_literal: true
+
+module Leva
+  # Controller for viewing the design system components
+  class DesignSystemController < ApplicationController
+    def index
+    end
+  end
+end

data/app/controllers/leva/experiments_controller.rb

@@ -4,9 +4,9 @@ module Leva
   class ExperimentsController < ApplicationController
     include ApplicationHelper
 
-    before_action :set_experiment, only: [:show, :edit, :update]
-    before_action :check_editable, only: [:edit, :update]
-    before_action :load_runners_and_evaluators, only: [:new, :edit, :create, :update]
+    before_action :set_experiment, only: [ :show, :edit, :update ]
+    before_action :check_editable, only: [ :edit, :update ]
+    before_action :load_runners_and_evaluators, only: [ :new, :edit, :create, :update ]
 
     # GET /experiments
     # @return [void]
@@ -39,7 +39,7 @@ module Leva
 
       if @experiment.save
         ExperimentJob.perform_later(@experiment) unless @experiment.completed?
-        redirect_to @experiment, notice: 'Experiment was successfully created and is now running.'
+        redirect_to @experiment, notice: "Experiment was successfully created and is now running."
       else
         render :new
       end
@@ -49,7 +49,7 @@ module Leva
     # @return [void]
     def update
       if @experiment.update(experiment_params)
-        redirect_to @experiment, notice: 'Experiment was successfully updated.'
+        redirect_to @experiment, notice: "Experiment was successfully updated."
       else
         render :edit
       end
@@ -69,7 +69,7 @@ module Leva
       # Queue the job again
       ExperimentJob.perform_later(@experiment)
 
-      redirect_to @experiment, notice: 'Experiment has been reset and is now running again.'
+      redirect_to @experiment, notice: "Experiment has been reset and is now running again."
     end
 
     private
@@ -92,7 +92,7 @@ module Leva
     end
 
     def check_editable
-      redirect_to @experiment, alert: 'Completed experiments cannot be edited.' if @experiment.completed?
+      redirect_to @experiment, alert: "Completed experiments cannot be edited." if @experiment.completed?
     end
   end
-end
+end

data/app/controllers/leva/runner_results_controller.rb

@@ -5,4 +5,4 @@ module Leva
       @runner_result = @experiment.runner_results.find(params[:id])
     end
   end
-end
+end

data/app/controllers/leva/workbench_controller.rb

@@ -4,9 +4,9 @@ module Leva
   class WorkbenchController < ApplicationController
     include ApplicationHelper
 
-    before_action :set_prompt, only: [:index, :edit, :update, :run, :run_all_evals, :run_evaluator]
-    before_action :set_dataset_record, only: [:index, :run, :run_all_evals, :run_evaluator]
-    before_action :set_runner_result, only: [:index, :run_all_evals, :run_evaluator]
+    before_action :set_prompt, only: [ :index, :edit, :update, :run, :run_all_evals, :run_evaluator ]
+    before_action :set_dataset_record, only: [ :index, :run, :run_all_evals, :run_evaluator ]
+    before_action :set_runner_result, only: [ :index, :run_all_evals, :run_evaluator ]
 
     # GET /workbench
     # @return [void]
@@ -17,6 +17,17 @@ module Leva
       @runners = load_runners
       @selected_runner = params[:runner] || @runners.first&.name
       @selected_dataset_record = params[:dataset_record_id] || DatasetRecord.first&.id
+
+      # Get merged context if runner and dataset record are available
+      if @selected_runner && @dataset_record
+        runner_class = @selected_runner.constantize rescue nil
+        if runner_class && runner_class < Leva::BaseRun
+          runner = runner_class.new
+          @record_context = @dataset_record.recordable.to_llm_context
+          @runner_context = runner.to_llm_context(@dataset_record.recordable)
+          @merged_context = @record_context.merge(@runner_context)
+        end
+      end
     end
 
     # GET /workbench/new
@@ -31,7 +42,7 @@ module Leva
     def create
       @prompt = Prompt.new(prompt_params)
       if @prompt.save
-        redirect_to workbench_index_path(prompt_id: @prompt.id), notice: 'Prompt was successfully created.'
+        redirect_to workbench_index_path(prompt_id: @prompt.id), notice: "Prompt was successfully created."
       else
         render :new
       end
@@ -47,45 +58,45 @@ module Leva
     def update
       @prompt = Prompt.find(params[:id])
       if @prompt.update(prompt_params)
-        render json: { status: 'success', message: 'Prompt updated successfully' }
+        render json: { status: "success", message: "Prompt updated successfully" }
       else
-        render json: { status: 'error', errors: @prompt.errors.full_messages }, status: :unprocessable_entity
+        render json: { status: "error", errors: @prompt.errors.full_messages }, status: :unprocessable_entity
       end
     end
 
     def run
-      return redirect_to workbench_index_path, alert: 'Please select a record and a runner' unless @dataset_record && run_params[:runner]
+      return redirect_to workbench_index_path, alert: "Please select a record and a runner" unless @dataset_record && run_params[:runner]
 
       runner_class = run_params[:runner].constantize
-      return redirect_to workbench_index_path, alert: 'Invalid runner selected' unless runner_class < Leva::BaseRun
+      return redirect_to workbench_index_path, alert: "Invalid runner selected" unless runner_class < Leva::BaseRun
 
       runner = runner_class.new
       runner_result = runner.execute_and_store(nil, @dataset_record, @prompt)
 
-      redirect_to workbench_index_path(prompt_id: @prompt.id, dataset_record_id: @dataset_record.id, runner: run_params[:runner]), notice: 'Run completed successfully'
+      redirect_to workbench_index_path(prompt_id: @prompt.id, dataset_record_id: @dataset_record.id, runner: run_params[:runner]), notice: "Run completed successfully"
     end
 
     def run_all_evals
-      return redirect_to workbench_index_path, alert: 'No runner result available' unless @runner_result
+      return redirect_to workbench_index_path, alert: "No runner result available" unless @runner_result
 
       load_evaluators.each do |evaluator_class|
         evaluator = evaluator_class.new
         evaluator.evaluate_and_store(nil, @runner_result)
       end
 
-      redirect_to workbench_index_path(prompt_id: @prompt.id, dataset_record_id: @dataset_record.id, runner: params[:runner]), notice: 'All evaluations completed successfully'
+      redirect_to workbench_index_path(prompt_id: @prompt.id, dataset_record_id: @dataset_record.id, runner: params[:runner]), notice: "All evaluations completed successfully"
     end
 
     def run_evaluator
-      return redirect_to workbench_index_path, alert: 'No runner result available' unless @runner_result
+      return redirect_to workbench_index_path, alert: "No runner result available" unless @runner_result
 
       evaluator_class = params[:evaluator].constantize
-      return redirect_to workbench_index_path, alert: 'Invalid evaluator selected' unless evaluator_class < Leva::BaseEval
+      return redirect_to workbench_index_path, alert: "Invalid evaluator selected" unless evaluator_class < Leva::BaseEval
 
       evaluator = evaluator_class.new
       evaluator.evaluate_and_store(nil, @runner_result)
 
-      redirect_to workbench_index_path(prompt_id: @prompt.id, dataset_record_id: @dataset_record.id, runner: params[:runner]), notice: 'Evaluator run successfully'
+      redirect_to workbench_index_path(prompt_id: @prompt.id, dataset_record_id: @dataset_record.id, runner: params[:runner]), notice: "Evaluator run successfully"
     end
 
     private
@@ -110,4 +121,4 @@ module Leva
       @runner_result = @dataset_record.runner_results.last if @dataset_record
     end
   end
-end
+end

data/app/helpers/leva/application_helper.rb

@@ -4,24 +4,24 @@ module Leva
     #
     # @return [Array<Class>] An array of evaluator classes
     def load_evaluators
-      load_classes_from_directory('app/evals', Leva::BaseEval) || []
+      load_classes_from_directory("app/evals", Leva::BaseEval) || []
     end
 
     # Loads all runner classes that inherit from Leva::BaseRun
     #
     # @return [Array<Class>] An array of runner classes
     def load_runners
-      load_classes_from_directory('app/runners', Leva::BaseRun) || []
+      load_classes_from_directory("app/runners", Leva::BaseRun) || []
     end
 
     # Loads predefined prompts from markdown files
     #
     # @return [Array<Array<String, String>>] An array of prompt name and content pairs
     def load_predefined_prompts
-      prompts = Dir.glob(Rails.root.join('app', 'prompts', '*.md')).map do |file|
-        name = File.basename(file, '.md').titleize
+      prompts = Dir.glob(Rails.root.join("app", "prompts", "*.md")).map do |file|
+        name = File.basename(file, ".md").titleize
         content = File.read(file)
-        [name, content]
+        [ name, content ]
       end
       prompts
     end
@@ -34,8 +34,8 @@ module Leva
     # @param base_class [Class] The base class that loaded classes should inherit from
     # @return [Array<Class>] An array of loaded classes
     def load_classes_from_directory(directory, base_class)
-      classes = Dir[Rails.root.join(directory, '*.rb')].map do |file|
-        File.basename(file, '.rb').camelize.constantize
+      classes = Dir[Rails.root.join(directory, "*.rb")].map do |file|
+        File.basename(file, ".rb").camelize.constantize
       end.select { |klass| klass < base_class }
       classes.empty? ? [] : classes
     end

data/app/jobs/leva/experiment_job.rb

@@ -18,4 +18,4 @@ module Leva
       end
     end
   end
-end
+end

data/app/jobs/leva/run_eval_job.rb

@@ -37,4 +37,4 @@ module Leva
       experiment.dataset.dataset_records.count == experiment.runner_results.count
     end
   end
-end
+end

data/app/models/concerns/leva/recordable.rb

@@ -3,10 +3,10 @@ module Leva
     extend ActiveSupport::Concern
 
     included do
-      has_many :dataset_records, as: :recordable, class_name: 'Leva::DatasetRecord', dependent: :destroy
-      has_many :datasets, through: :dataset_records, class_name: 'Leva::Dataset'
-      has_many :runner_results, through: :dataset_records, class_name: 'Leva::RunnerResult'
-      has_many :evaluation_results, through: :runner_results, class_name: 'Leva::EvaluationResult'
+      has_many :dataset_records, as: :recordable, class_name: "Leva::DatasetRecord", dependent: :destroy
+      has_many :datasets, through: :dataset_records, class_name: "Leva::Dataset"
+      has_many :runner_results, through: :dataset_records, class_name: "Leva::RunnerResult"
+      has_many :evaluation_results, through: :runner_results, class_name: "Leva::EvaluationResult"
     end
 
     # @return [String] The ground truth label for the record
@@ -34,4 +34,4 @@ module Leva
       false
     end
   end
-end
+end

data/app/models/leva/dataset.rb

@@ -25,4 +25,4 @@ module Leva
       end
     end
   end
-end
+end

data/app/models/leva/evaluation_result.rb

@@ -36,4 +36,4 @@ module Leva
 
     scope :for_evaluator, ->(evaluator_class) { where(evaluator_class: evaluator_class.name) }
   end
-end
+end

data/app/models/leva/experiment.rb

@@ -40,4 +40,4 @@ module Leva
 
     serialize :evaluator_classes, coder: JSON, type: Array
   end
-end
+end

data/app/models/leva/prompt.rb

@@ -27,4 +27,4 @@ module Leva
       self.version += 1
     end
   end
-end
+end

data/app/views/layouts/leva/application.html.erb

@@ -1,40 +1,39 @@
 <!DOCTYPE html>
-<html lang="en" class="bg-gray-950">
+<html lang="en">
   <head>
     <meta charset="UTF-8">
     <meta name="viewport" content="width=device-width, initial-scale=1.0">
     <title>Leva - <%= yield(:title) || 'AI Evaluation Engine' %></title>
     <%= csrf_meta_tags %>
     <%= csp_meta_tag %>
-    <script src="https://cdn.tailwindcss.com"></script>
+
+    <%# Fira Code for monospace, Helvetica is system font %>
+    <link rel="preconnect" href="https://fonts.googleapis.com">
+    <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+    <link href="https://fonts.googleapis.com/css2?family=Fira+Code:wght@400;500&display=swap" rel="stylesheet">
+
+    <%= stylesheet_link_tag "leva/application", media: "all" %>
     <script src="https://cdn.jsdelivr.net/npm/stimulus@3.2.2/dist/stimulus.umd.min.js"></script>
     <%= yield(:head) %>
   </head>
-  <body class="bg-gray-950 text-white">
-    <nav class="bg-gray-900 border-b border-gray-800">
-      <div class="px-4">
-        <div class="flex h-16 items-center">
-          <%= link_to root_path, class: 'flex items-center text-2xl font-bold text-indigo-400' do %>
-            <svg xmlns="http://www.w3.org/2000/svg" class="h-8 w-8 mr-2" viewBox="0 0 24 24" fill="none" stroke="url(#gradient)" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
-              <defs>
-                <linearGradient id="gradient" x1="0%" y1="0%" x2="100%" y2="100%">
-                  <stop offset="0%" stop-color="#4F46E5" />
-                  <stop offset="100%" stop-color="#9333EA" />
-                </linearGradient>
-              </defs>
-              <path d="M12 2L2 7l10 5 10-5-10-5zM2 17l10 5 10-5M2 12l10 5 10-5" />
-            </svg>
-            LLM Evals
-          <% end %>
-          <div class="ml-10 flex space-x-6">
-            <%= link_to 'Workbench', workbench_index_path, class: "px-3 py-2 rounded-md text-sm font-medium transition-colors duration-150 ease-in-out #{request.path.start_with?(workbench_index_path) ? 'bg-indigo-600 text-white shadow-lg' : 'text-gray-300 hover:bg-gray-800 hover:text-white'}" %>
-            <%= link_to 'Datasets', datasets_path, class: "px-3 py-2 rounded-md text-sm font-medium transition-colors duration-150 ease-in-out #{request.path.start_with?(datasets_path) ? 'bg-indigo-600 text-white shadow-lg' : 'text-gray-300 hover:bg-gray-800 hover:text-white'}" %>
-            <%= link_to 'Experiments', experiments_path, class: "px-3 py-2 rounded-md text-sm font-medium transition-colors duration-150 ease-in-out #{request.path.start_with?(experiments_path) ? 'bg-indigo-600 text-white shadow-lg' : 'text-gray-300 hover:bg-gray-800 hover:text-white'}" %>
-          </div>
+  <body>
+    <nav class="nav">
+      <div class="nav-inner">
+        <%= link_to root_path, class: 'nav-brand' do %>
+          <svg viewBox="0 0 20 20" fill="none" xmlns="http://www.w3.org/2000/svg" class="nav-logo">
+            <rect x="2" y="2" width="16" height="16" rx="3" fill="currentColor" fill-opacity="0.15"/>
+            <path d="M6 6V14H14" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round"/>
+          </svg>
+          <span>leva</span>
+        <% end %>
+        <div class="nav-links">
+          <%= link_to 'Workbench', workbench_index_path, class: "nav-link #{request.path == root_path || request.path.start_with?(workbench_index_path) ? 'active' : ''}" %>
+          <%= link_to 'Datasets', datasets_path, class: "nav-link #{request.path.start_with?(datasets_path) ? 'active' : ''}" %>
+          <%= link_to 'Experiments', experiments_path, class: "nav-link #{request.path.start_with?(experiments_path) ? 'active' : ''}" %>
         </div>
       </div>
     </nav>
-    <main class="bg-gray-950">
+    <main>
       <%= yield %>
     </main>
   </body>

data/app/views/leva/dataset_records/index.html.erb

@@ -1,49 +1,76 @@
 <% content_for :title, "#{@dataset.name} - Records" %>
-<div class="container mx-auto px-4 py-8 bg-gray-950 text-white">
-  <div class="mb-8">
-    <div class="flex justify-between items-center">
-      <h1 class="text-3xl font-bold text-indigo-400 mb-2"><%= @dataset.name %> - Records</h1>
-      <%= link_to dataset_path(@dataset), class: 'btn btn-secondary flex items-center' do %>
-        <svg xmlns="http://www.w3.org/2000/svg" class="h-5 w-5 mr-2" viewBox="0 0 20 20" fill="currentColor">
-          <path fill-rule="evenodd" d="M9.707 16.707a1 1 0 01-1.414 0l-6-6a1 1 0 010-1.414l6-6a1 1 0 011.414 1.414L5.414 9H17a1 1 0 110 2H5.414l4.293 4.293a1 1 0 010 1.414z" clip-rule="evenodd" />
-        </svg>
-        Back to Dataset
-      <% end %>
+<div class="container page">
+  <%# Breadcrumb Navigation %>
+  <nav class="breadcrumb mb-4">
+    <%= link_to datasets_path, class: 'breadcrumb-link' do %>Datasets<% end %>
+    <span class="breadcrumb-sep">/</span>
+    <%= link_to @dataset.name, dataset_path(@dataset), class: 'breadcrumb-link' %>
+    <span class="breadcrumb-sep">/</span>
+    <span class="breadcrumb-current">Records</span>
+  </nav>
+
+  <div class="page-header">
+    <div>
+      <h1 class="page-title">Dataset Records</h1>
+      <p class="text-muted mt-2">All records in <strong><%= @dataset.name %></strong></p>
     </div>
   </div>
 
-  <div class="bg-gray-800 rounded-lg shadow-lg overflow-hidden">
-    <table class="min-w-full divide-y divide-gray-700">
-      <thead class="bg-gray-700">
-        <tr>
-          <% @records.first.index_attributes.keys.each do |key| %>
-            <th scope="col" class="px-6 py-3 text-left text-xs font-medium text-gray-300 uppercase tracking-wider">
-              <%= key.to_s.humanize %>
-            </th>
-          <% end %>
-          <th scope="col" class="relative px-6 py-3">
-            <span class="sr-only">Actions</span>
-          </th>
-        </tr>
-      </thead>
-      <tbody class="bg-gray-800 divide-y divide-gray-700">
-        <% @records.each do |record| %>
-          <tr class="hover:bg-gray-700 transition-colors duration-200">
-            <% record.index_attributes.values.each do |value| %>
-              <td class="px-6 py-4 whitespace-nowrap text-sm text-gray-300">
-                <%= value %>
-              </td>
+  <% if @records.any? %>
+    <%# Section Header with Count %>
+    <div class="section-header">
+      <h3 class="section-title">Records</h3>
+      <span class="section-count"><%= @records.count %> total</span>
+    </div>
+
+    <div class="table-wrapper">
+      <div class="table-scroll">
+        <table class="table table-clickable">
+          <thead>
+            <tr>
+              <% @records.first.index_attributes.keys.each do |key| %>
+                <th><%= key.to_s.humanize %></th>
+              <% end %>
+              <th class="text-right" style="width: 50px;"></th>
+            </tr>
+          </thead>
+          <tbody>
+            <% @records.each do |record| %>
+              <tr class="clickable-row" onclick="window.location='<%= dataset_dataset_record_path(@dataset, record) %>'">
+                <% record.index_attributes.each_with_index do |(key, value), index| %>
+                  <td class="<%= index == 0 ? '' : 'cell-truncate' %>">
+                    <%= index == 0 ? content_tag(:span, value, class: 'row-title') : value %>
+                  </td>
+                <% end %>
+                <td class="text-right">
+                  <svg class="icon-sm text-muted" viewBox="0 0 20 20" fill="currentColor" style="opacity: 0.5;">
+                    <path fill-rule="evenodd" d="M7.293 14.707a1 1 0 010-1.414L10.586 10 7.293 6.707a1 1 0 011.414-1.414l4 4a1 1 0 010 1.414l-4 4a1 1 0 01-1.414 0z" clip-rule="evenodd" />
+                  </svg>
+                </td>
+              </tr>
             <% end %>
-            <td class="px-6 py-4 whitespace-nowrap text-right text-sm font-medium">
-              <%= link_to 'View', dataset_dataset_record_path(@dataset, record), class: 'text-indigo-400 hover:text-indigo-300 transition-colors duration-200' %>
-            </td>
-          </tr>
-        <% end %>
-      </tbody>
-    </table>
-  </div>
+          </tbody>
+        </table>
+      </div>
+    </div>
+  <% else %>
+    <div class="card p-6">
+      <div class="empty-state" style="padding: var(--space-10) var(--space-6);">
+        <svg class="empty-state-icon" fill="none" viewBox="0 0 24 24" stroke="currentColor">
+          <path stroke-linecap="round" stroke-linejoin="round" stroke-width="1.5" d="M19 11H5m14 0a2 2 0 012 2v6a2 2 0 01-2 2H5a2 2 0 01-2-2v-6a2 2 0 012-2m14 0V9a2 2 0 00-2-2M5 11V9a2 2 0 012-2m0 0V5a2 2 0 012-2h6a2 2 0 012 2v2M7 7h10" />
+        </svg>
+        <h3 class="empty-state-title">No records in this dataset</h3>
+        <p class="empty-state-description">Add records to start building your evaluation dataset.</p>
+      </div>
 
-  <div class="mt-4 text-gray-400 text-sm">
-    Total records: <%= @records.count %>
-  </div>
-</div>
+      <div class="divider"></div>
+
+      <div class="p-5" style="background: var(--gray-850, #222120); border-radius: var(--radius-md);">
+        <h4 class="text-sm font-semibold mb-3" style="color: var(--gray-300);">Adding Records</h4>
+        <p class="text-sm text-muted mb-3">Use the Rails console to add records to this dataset:</p>
+        <pre style="margin: 0;"><code>dataset = Leva::Dataset.find(<%= @dataset.id %>)
+dataset.add_record(your_model_instance)</code></pre>
+      </div>
+    </div>
+  <% end %>
+</div>