AI Chatbot Platform

🚀 Key Features

🤖 AI-Powered Conversations

• Multi-Provider Support: OpenAI GPT-4 & Google Gemini integration

• Streaming Responses: Real-time AI responses with Server-Sent Events (SSE)

• Smart Context Management: Maintains conversation history and context

• Cost Tracking: Monitor AI usage and token consumption

• Retry Logic: Automatic retry with exponential backoff for reliability

• Timeout Handling: 30-second timeout protection with graceful fallbacks

👥 Live Operator System

• AI-to-Operator Handoff: Seamless transition from AI to human operator

• Real-Time Dashboard: Live operator interface with active conversations

• Priority Queue System: 4-level priority (low/normal/high/urgent)

• Auto-Assignment: 6 intelligent strategies (Longest Wait, Lowest Workload, Round Robin, Skill-Based, Hybrid, Random)

• Queue Position Tracking: Real-time position updates with estimated wait times for users

• Session Tracking: Performance metrics per operator

• Internal Notes & Tags: Conversation management tools

• Status Management: Online/Away/Busy/Offline operator states

• Transfer Capability: Move conversations between operators

• Queue Analytics: Wait time, resolution time, satisfaction metrics

• Admin Queue Configuration: Configurable assignment strategies with testing capabilities

💬 Embeddable Widget

• One-Line Integration: Simple <script> tag embedding

• Fully Customizable: Colors, logo, position, branding

• Pre-Chat Forms: Collect user information before chat starts

• Domain Security: Whitelist-based CORS protection

• Mobile Responsive: Works seamlessly on all devices

• Sound Notifications: Optional audio alerts

• Minimized/Maximized States: User-friendly interface

• Public Token Auth: No user login required for visitors

• Hierarchical Action Buttons: WhatsApp-like service navigation system

• 3 Display Styles: Buttons, List, or Cards for action buttons

• Unlimited Nesting Depth: Service → Sub-Service → Content hierarchy

• Click Analytics: Track user navigation patterns through action buttons

• Powered By Branding: Configurable "Powered by" link with theming

• Security Alerts: Visual warnings for policy violations (strike system)

• Operator Request Cooldown: 10-second cooldown to prevent spam

• Keyword Detection: Automatic operator detection (e.g., "operator", "canlı destek")

📚 Knowledge Base (RAG)

• Document Upload: Support for PDF, DOCX, DOC (with warning), TXT, MD formats

• Enhanced PDF Extraction: Quality assessment, better text cleaning, detailed error messages

• Improved DOCX Processing: Better structure preservation, table handling, extraction warnings

• Smart Text Cleaning: RAG-optimized cleaning that preserves meaningful structure

• Validation & Error Handling: File integrity checks, helpful error messages, quality metrics

• 3 Chunking Strategies: Recursive, Fixed-size, Semantic

• Multi-Provider Embeddings: Gemini (free/multilingual), OpenAI (cloud/paid), HuggingFace (local)

• Provider Switching: Zero-code switching via admin panel configuration

• 6 HuggingFace Models: all-MiniLM-L6-v2, all-mpnet-base-v2, BGE variants

• Hybrid Search: Vector similarity + keyword BM25 search

• Semantic Retrieval: Context-aware document search with RAG

• Real-Time Streaming: RAG context integrated with streaming responses

• Document Management: Upload, view, reindex, delete operations

• Usage Analytics: Track retrieval effectiveness with quality scores

• MinIO Storage: Scalable object storage backend

• UTF-8 Support: Full multi-byte character handling (Azerbaijani, Turkish, etc.)

🚀 Enterprise RAG Pipeline (Advanced)

• Query Rewrite Layer: AI-powered query optimization before vector search

  • Automatic typo correction and normalization

  • Query expansion with synonyms

  • Separate AI model (gpt-4o-mini) to avoid impacting main responses

  • Smart caching to reduce API calls and costs

• Reranking Layer: Cross-encoder reranking for improved accuracy

  • Fetch more candidates (topN), rerank to best matches (topK)

  • Configurable truncate length (500-5000 chars) for context optimization

  • LLM-based scoring with relevance explanations

• Answer Validation Layer: Verify AI responses against source documents

  • Hallucination detection with PASS/FAIL/UNCERTAIN verdicts

  • Automatic regeneration on validation failure

  • Confidence scoring with detailed reasoning

• Metadata Filtering: Filter documents by tags, categories, dates

  • Widget-specific document access control

  • Dynamic filter building from query context

• Legal Text Preservation: Special handling for legal documents

  • Preserve article boundaries (Maddə, Bənd, Hissə, Fəsil)

  • Configurable heading patterns for different document types

• Query Normalization: Multi-language text preprocessing

  • Azerbaijani/Turkish special character handling

  • Automatic language detection

  • Search variant generation for better matching

• RAG Caching: Intelligent caching for cost optimization

  • Configurable TTL (60-3600 seconds)

  • Cache statistics and manual clear functionality

  • Separate caches for query rewrite and reranking

💰 Cost Optimization Features

• Smart Model Selection: Use cheaper models (gpt-4o-mini) for RAG processing

• Caching System: Reduce duplicate API calls with configurable TTL caching

• Truncation Control: Limit context size to reduce token usage

• Confidence-Based Processing: Only apply expensive RAG layers when needed

• Free Embeddings: Gemini embedding models with zero cost

• Fallback Providers: Automatic failover to backup AI providers

• Token Monitoring: Real-time tracking of API usage and costs

• Admin Dashboard: Visual cost analytics and efficiency metrics