ai-support

AI Support Assistant

RAG-powered customer support system demonstrating ObjectStack AI protocols

Overview

This example demonstrates how to build an intelligent customer support system using ObjectStack's AI protocols:

• RAG Pipeline: Knowledge base powered by vector search
• AI Agent: Intelligent support agent with function calling
• Model Registry: Centralized LLM configuration with fallback support
• Natural Language Queries: Business users can query tickets using natural language

Features

1. RAG-Powered Knowledge Base

• Vector search across documentation
• Semantic chunking of markdown content
• MMR retrieval for diverse results
• Cohere reranking for accuracy

2. Intelligent Support Agent

• Answer customer questions using RAG
• Create and update support tickets
• Search existing tickets for similar issues
• Escalate to human agents when needed
• Collect customer satisfaction feedback

3. Natural Language Ticketing

• "Show me all high priority tickets"
• "What tickets are assigned to me?"
• "How many open tickets from last week?"

Architecture

┌─────────────────────────────────────────────────────────┐
│                   Customer Interface                     │
│              (Chat, Email, Web Portal)                   │
└──────────────────────┬──────────────────────────────────┘
                       │
┌──────────────────────▼──────────────────────────────────┐
│                  AI Support Agent                        │
│           (GPT-4 Turbo with function calling)            │
└──────────────────────┬──────────────────────────────────┘
                       │
        ┌──────────────┼──────────────┐
        │              │              │
┌───────▼──────┐ ┌────▼─────┐ ┌─────▼──────┐
│ RAG Pipeline │ │  Actions  │ │  ObjectQL  │
│  (Pinecone)  │ │  (CRUD)   │ │  (Query)   │
└──────────────┘ └───────────┘ └────────────┘

Configuration Files

• src/ai-config.ts - AI agent, model registry, and RAG pipeline
• src/domains/ticket.object.ts - Support ticket data model
• objectstack.config.ts - Application configuration

Quick Start

# Install dependencies
pnpm install

# Start development server
pnpm dev

AI Configuration

Agent Configuration

export const SupportAgent: Agent = {
  name: 'customer_support_ai',
  label: 'AI Support Agent',
  role: 'Senior Customer Support Specialist',
  
  tools: [
    { type: 'action', name: 'create_support_ticket' },
    { type: 'vector_search', name: 'kb_search' },
    { type: 'query', name: 'search_tickets' },
  ],
  
  knowledge: {
    topics: ['product_documentation', 'faq', 'troubleshooting'],
    indexes: ['support_kb_v2'],
  },
};

RAG Pipeline

export const KnowledgeBaseRAG: RAGPipelineConfig = {
  embedding: {
    provider: 'openai',
    model: 'text-embedding-3-large',
    dimensions: 3072,
  },
  
  vectorStore: {
    provider: 'pinecone',
    indexName: 'support-kb-prod',
  },
  
  retrieval: {
    type: 'mmr',
    topK: 5,
    lambda: 0.7,
  },
};

Example Interactions

Customer Query

Customer: "My API calls are failing with a 401 error"

AI Agent:

1. Searches knowledge base for authentication errors
2. Finds relevant documentation
3. Provides step-by-step troubleshooting
4. If unresolved, creates a ticket and escalates

Natural Language Ticket Query

Support Manager: "Show me all critical tickets from the last 24 hours"

System:

1. Parses natural language to ObjectQL
2. Executes query: ticket.filter(priority='critical', created_date>yesterday)
3. Returns formatted results with summaries

Success Metrics

• Response Time: < 2s for RAG queries
• Accuracy: 90%+ customer satisfaction
• Automation: 70% of tickets handled without human intervention
• Knowledge Coverage: 95% of queries have relevant KB articles

Related Examples

• AI Data Analyst - NLQ for business intelligence
• AI Code Generator - Generate ObjectStack apps from descriptions
• AI Sales Assistant - Intelligent sales automation