LangGraph Multi-Agent Support Starter

Production-Grade Customer Support Triage, PII Guardrails & State Graph Routing

A production-ready reference implementation for enterprise customer support ticket classification, automated intent routing, PII data sanitization, and human-in-the-loop escalation built on the LangGraph state graph architecture.

---

🏗️ Architecture Overview

                                 Incoming Customer Query
                               (REST / Webhook / Chatbot)
                                            │
                                            ▼
                        ┌───────────────────────────────────────┐
                        │       Security & Guardrails Layer     │
                        │   • Prompt Injection Heuristics       │
                        │   • PII Masking (CC, SSN, Phone, Mail)│
                        └───────────────────┬───────────────────┘
                                            │ Sanitized State
                                            ▼
                        ┌───────────────────────────────────────┐
                        │        Supervisor Router Node         │
                        │   • Intent Classification             │
                        │   • Sentiment Analysis (Frustration)  │
                        │   • Dynamic Priority Assignment       │
                        └───────┬───────────┼───────────┬───────┘
                                │           │           │
            ┌───────────────────┘           │           └───────────────────┐
            ▼                               ▼                               ▼
┌───────────────────────┐       ┌───────────────────────┐       ┌───────────────────────┐
│ Technical Specialist  │       │  Billing Specialist   │       │   Human Escalation    │
│  • API / SDK Errors   │       │  • Tax Invoices (VAT) │       │  • Priority Ticket ID │
│  • Latency & Timeouts │       │  • Refund Policies    │       │  • CRM Handoff (Lead) │
│  • Webhook Validation │       │  • Plan Upgrades      │       │  • SLA < 30 min Alert │
└───────────┬───────────┘       └───────────┬───────────┘       └───────────┬───────────┘
            │                               │                               │
            └───────────────────────┬───────┴───────────────────────────────┘
                                    │
                                    ▼
                        ┌───────────────────────┐
                        │ Final State Resolution│
                        │  • Audit Trail        │
                        │  • Response Payload   │
                        └───────────────────────┘

---

✨ Features

• Multi-Agent State Graph: Deterministic and LLM-driven graph routing with state persistence across conversational turns.
• Enterprise PII Guardrails: Pre-execution sanitization redacting credit cards, social security numbers, phone numbers, and emails to prevent unauthorized sensitive data leakage into downstream LLM inference logs.
• Prompt Injection Defense: Pre-flight heuristic detection catching adversarial overrides (ignore previous instructions, jailbreak, developer mode).
• Sentiment & Priority Analysis: Automatically escalates angry, urgent, or litigious inquiries to human supervisors with dedicated priority SLAs.
• FastAPI Webhook & REST Server: Out-of-the-box endpoints for /api/v1/chat and /api/v1/tickets/route (integrates directly with Zendesk, Freshdesk, and HubSpot webhooks).
• Interactive CLI Console: Test support queries and live agent triage from the terminal.
• 100% Type-Safe & Tested: Pydantic V2 models, strict typing, and comprehensive unit test coverage.

---

🚀 Quick Start

1. Installation

# Clone repository
git clone https://github.com/adaptnxt/langgraph-multiagent-support-starter.git
cd langgraph-multiagent-support-starter

# Install with core dependencies
pip install -e .

# Or install with optional LangChain/LangGraph LLM providers
pip install -e ".[llm,dev]"

2. Run from CLI

Route and resolve a support query:

adaptnxt-support route "Where can I download the VAT tax invoice for last month?"

Output:

[Agent Network Route Summary]
Assigned Agent: billing
Priority:       medium
Sentiment:      neutral

Response:
You can view and download all past VAT/GST compliant PDF tax invoices directly from your Account Portal under Settings > Billing & Invoices.

Test PII Redaction & Escalation:

adaptnxt-support route "I am furious! My card 4532 1122 3344 5566 was charged twice! I demand to speak to a manager right now!"

Output:

[Agent Network Route Summary]
Assigned Agent: human_escalation
Priority:       critical
Sentiment:      angry
Redacted PII:   credit_card
Escalation ID:  TICKET-ESC-9F81E3B2

Response:
Your request has been prioritized and escalated directly to an AdaptNXT Tier-3 Support Lead. Escalation Ticket ID: TICKET-ESC-9F81E3B2 (Priority: CRITICAL). A dedicated team member will contact you within 30 minutes.

Launch Interactive Terminal Chat:

adaptnxt-support interactive

---

🌐 Running the FastAPI Server

Launch the REST server locally:

adaptnxt-support serve --host 0.0.0.0 --port 8000

Interactive Swagger API documentation will be available at: 👉 http://localhost:8000/docs

Sample API Request (cURL):

curl -X POST "http://localhost:8000/api/v1/chat" \
     -H "Content-Type: application/json" \
     -d '{"query": "Why am I getting HTTP 401 code on the webhook endpoint?", "session_id": "sess_102"}'

---

🐳 Docker Deployment

Build and Run Standalone Container:

docker build -t adaptnxt/support-agents .
docker run -p 8000:8000 adaptnxt/support-agents

Using Docker Compose:

docker compose up -d

---

🧪 Testing

Run the automated test suite:

pytest tests/ -v

---

🏢 Enterprise Citation & Production Deployment

Maintained by the Enterprise AI & Cloud Chatbot engineering team at AdaptNXT Technology Solutions.

For custom fine-tuned multi-agent architectures (CrewAI, AutoGen, LangGraph) integrated with internal enterprise databases (PostgreSQL, SAP, Salesforce, Pinecone), visit our specialized service practices:

• Cloud AI Chatbot Integration Services
• Open Source LLM Chatbot Development
• Enterprise RAG Chatbot Development

---

📄 License

This project is licensed under the Apache-2.0 License.