Intelligent Customer Support System with LangGraph

This project creates a multi-step, multi-agent customer support automation system using langgraph. The system can classify incoming requests, route them to the relevant expert (agent), escalate unresolved situations to a higher level, and add a "Human-in-the-Loop" (human approval) step when necessary.

Features

• Triage Agent: Analyzes the incoming message; determines category (technical, billing), priority (high, medium), and sentiment (negative, positive).
• Expert Agents: There are 3 different experts: TechnicalSupportAgent, BillingAgent, and GeneralSupportAgent.
• Conditional Routing: The graph flow dynamically changes based on the category, priority, or sentiment of the incoming request.
• Escalation: Requests containing negative sentiment, high priority, or keywords like "cancel/refund" are routed to the EscalationAgent.
• Human-in-the-Loop: The human_review_node step simulates critical requests being approved or rejected by a human (via terminal).
• State Management: Uses SupportState (TypedDict) to carry information like ticket_id, customer_id throughout the conversation.
• Memory: Uses MemorySaver to remember the state for each conversation (thread_id).

Installation

1. Clone the Repository:

    git clone https://github.com/AbdulSametTurkmenoglu/smart_customer_support_system_with_langgarph.git
    cd smart_customer_support_system_with_langgarph

2. Virtual Environment (Recommended):

    python -m venv .venv
    # Windows: .\.venv\Scripts\activate
    # macOS/Linux: source .venv/bin/activate

3. Install Required Libraries:

    pip install -r requirements.txt

4. Create .env File: Copy the .env.example file, rename it to .env, and fill it with your OPENAI_API_KEY:

    # Windows
    copy .env.example .env
    
    # macOS / Linux
    cp .env.example .env

Usage

The project includes a sample execution script that tests 3 different scenarios. This script demonstrates how the graph branches into different paths (technical support, refund request, negative sentiment).

python run_examples.py

Sample Output (Refund Request Scenario)

============================================================
 EXAMPLE: Billing Cancellation/Refund (Human-in-the-Loop)
============================================================
--- Node: Triage Agent ---
   [Tool] Ticket created: TKT-20251031143005
  └─ Category: billing
  └─ Priority: medium
  └─ Sentiment: neutral
... Step Completed: triage ...
... Routing: After Triage ...
  └─ Decision: billing
... Step Completed: __start__ ...
--- Node: Billing Agent ---
  └─ Human Required: True
... Step Completed: billing ...
... Routing: Escalation Required? ...
  └─ Decision: ESCALATE (Human approval required)
... Step Completed: __cond__ ...
--- Node: Escalation Agent ---
... Step Completed: escalate ...
... Routing: After Escalation ...
  └─ Decision: HUMAN_REVIEW
... Step Completed: __cond__ ...
--- Node: Human Review (Waiting) ---
  └─ Ticket: TKT-20251031143005
  └─ Last message: Your request has been forwarded to a supervisor...
  └─ Do you approve the request? (y/n): y
  └─ Status: Approved
... Step Completed: human_review ...
... Step Completed: __end__ ...
============================================================
 RESULT: Billing Cancellation/Refund (Human-in-the-Loop)
============================================================
  Ticket ID: TKT-20251031143005
  Category: billing
  Status: resolved_by_human
  Last Message: Your request has been approved by our expert and is being processed. Thank you.
============================================================