[v0.6.0] Smart Suggestions System

[abdullah4tech]

Description

Implement AI-powered suggestions and recommendations based on document content, query patterns, and user behavior to enhance the interactive experience.

User Stories

• As a user, I want to see suggested follow-up questions after each response
• As a user, I want recommendations for related topics I might want to explore
• As a user, I want quick action templates for common queries
• As a user, I want to discover content in my documents I haven't explored yet
• As a user, I want context-aware suggestions that understand my conversation flow

Tasks

Suggestion Components

• Create src/components/SmartSuggestions.vue

  • Suggestion cards container
  • Individual suggestion items
  • Click to auto-fill query
  • Refresh suggestions button
  • Dismiss/hide functionality
  • Loading state while generating
  • Empty state when no suggestions

• Create src/components/QuickActions.vue

  • Template library display
  • Template categories
  • Search templates
  • Click to apply template
  • Custom template creation

Store Management

• Create src/store/suggestions.ts
  • State for current suggestions
  • State for suggestion history
  • State for user preferences
  • Actions to fetch suggestions
  • Actions to track suggestion usage
  • Getters for filtered suggestions
  • Cache management

AI-Powered Suggestions

• Implement context analysis
  • Analyze recent conversation messages
  • Extract key topics and entities
  • Identify question patterns
  • Detect user intent
• Generate follow-up questions
  • Based on last AI response
  • Consider document sources cited
  • Explore related concepts
  • Drill deeper into topics
  • Compare/contrast suggestions
• Create query templates
  • "Summarize [document/section]"
  • "Compare [concept A] and [concept B]"
  • "What are the key points about [topic]?"
  • "Explain [term] in simple terms"
  • "Show examples of [concept]"
  • "What are the implications of [finding]?"

Document-Based Recommendations

• Extract key topics from documents
  • Use QuerySource data
  • Identify main themes
  • Extract section headings
  • Find frequently mentioned terms
• Suggest unexplored content
  • Identify unqueried documents
  • Find unvisited pages/sections
  • Highlight new uploads
  • Suggest related documents
• Generate "You might want to ask about..."
  • Based on document content
  • Based on similar user queries (if available)
  • Based on conversation context
  • Based on bookmarked content

Quick Actions System

• Create quick action templates

  export interface QuickAction {
    id: string
    name: string
    description: string
    category: string
    template: string
    variables?: string[] // Placeholders in template
  }

• Implement template categories

  • "Summarization" - summarize documents/sections
  • "Analysis" - analyze concepts/patterns
  • "Comparison" - compare items
  • "Explanation" - explain terms/concepts
  • "Examples" - request examples
  • "Custom" - user-created templates

• Template variable system

  • Identify placeholders (e.g., {document}, {topic})
  • Prompt user for variable values
  • Auto-fill from context when possible
  • Save filled templates for reuse

Backend Integration

• Update src/services/api.ts

  • Add getSuggestions() endpoint
  • Add getRelatedTopics() endpoint
  • Add getDocumentInsights() endpoint
  • Implement caching strategy
  • Error handling for failed suggestions

• Backend API Requirements (for backend team)

  // POST /api/suggestions
  interface SuggestionsRequest {
    conversationContext: Message[]
    currentDocument?: string
    queryHistory?: string[]
    maxSuggestions?: number
  }
  
  interface SuggestionsResponse {
    followUpQuestions: string[]
    relatedTopics: string[]
    unexploredContent: {
      documentId: string
      title: string
      reason: string
    }[]
    quickActions: QuickAction[]
  }

Integration with Existing Features

• Display suggestions in ChatView.vue
  • Below last AI message
  • As floating sidebar panel
  • Configurable position
• Use analytics data from v0.5.0
  • Suggest based on popular queries
  • Identify gaps in document coverage
  • Personalize based on usage patterns
• Integrate with bookmarks from v0.3.0
  • Suggest reviewing bookmarked content
  • Generate questions from bookmarks
  • Find related content to bookmarks

User Preferences

• Create suggestion settings
  • Enable/disable suggestions
  • Number of suggestions to show (3-10)
  • Auto-refresh frequency
  • Preferred suggestion types
  • Hide dismissed suggestions

Acceptance Criteria

• Suggestions appear after each AI response (if enabled)
• At least 3-5 relevant suggestions generated per response
• Suggestions are contextually relevant (>80% relevance)
• Can click suggestion to auto-fill query input
• Can dismiss individual suggestions
• Can refresh suggestions manually
• Quick action templates work correctly with variables
• Can create custom quick action templates
• Document recommendations highlight unexplored content
• Suggestions don't impact chat performance
• Suggestions load within 2 seconds
• Settings allow full customization of suggestion behavior
• Suggestion usage is tracked for future improvements

Dependencies

• All previous versions (v0.1.0 - v0.5.0)
• ragAPI.queryDocuments from src/services/api.ts
• Analytics data from v0.5.0
• Bookmark data from v0.3.0
• New backend endpoint required: /api/suggestions

Design Notes

Suggestion Display

• Show 3-5 suggestions by default
• Use card-based layout
• Each suggestion has:
  • Icon indicating type (question, document, action)
  • Suggestion text
  • Click to use
  • Dismiss button (small X)
• Subtle animation on appearance
• Horizontal scrolling on mobile

Quick Actions

• Grid layout for templates
• Category tabs at top
• Search bar for filtering
• Variable inputs as inline forms
• Preview of filled template

Colors & Icons

• Follow-up questions: 💡 (lightbulb) - Blue
• Related topics: 🔗 (link) - Purple
• Unexplored content: 📄 (document) - Green
• Quick actions: ⚡ (lightning) - Orange

Technical Considerations

Performance

• Cache suggestions for 5 minutes
• Debounce suggestion refresh (1 second)
• Load suggestions asynchronously
• Don't block chat functionality
• Implement request cancellation for rapid context changes

Suggestion Quality

• Use last 3-5 messages for context
• Limit context size to 2000 tokens
• Filter out repetitive suggestions
• Track dismissed suggestions to avoid re-suggesting
• A/B test different suggestion algorithms

Fallback Strategy

• If backend unavailable, use rule-based suggestions
• Extract nouns/entities from last response
• Generate "Tell me more about [entity]" suggestions
• Suggest document-based queries from metadata

Data Privacy

• Don't send sensitive conversation data to external APIs
• Process suggestions server-side when possible
• Allow users to opt-out completely
• Clear suggestion history on logout

Testing Checklist

• Test suggestions with short conversations (1-2 messages)
• Test suggestions with long conversations (50+ messages)
• Test with no conversation context (first message)
• Test quick action templates with variables
• Test creating custom quick actions
• Test suggestion relevance (manual review)
• Test suggestion performance (load time)
• Test with backend unavailable (fallback)
• Test dismissing suggestions
• Test refreshing suggestions
• Test settings toggle
• Test mobile layout
• Test with different document types
• Verify no duplicate suggestions

Related Issues

• Depends on: All previous versions (v0.1.0 - v0.5.0)
• Uses data from: [v0.3.0] Bookmarks & Labels System #4 (v0.5.0 Analytics)
• Uses data from: Setup Vitest for Testing Infrastructure #2 (v0.3.0 Bookmarks)
• Requires: Backend API enhancement

Milestone

v0.6.0 Release

Backend Requirements Checklist

• Create /api/suggestions endpoint
• Implement suggestion generation algorithm
• Add caching layer for suggestions
• Create suggestion relevance scoring
• Add rate limiting (prevent abuse)
• Document API schema
• Provide example responses
• Add monitoring/logging for suggestion quality

Additional Resources

• Consider using LangChain for suggestion generation
• Research: Query suggestion best practices
• Reference: Google search suggestions algorithm
• Reference: Amazon product recommendations system