🛠️ Support Ticket Resolution Agent with Multi-Step Review Loop

This project is an AI-powered support ticket agent built with LangGraph, Azure OpenAI, and FAISS + Hugging Face Transformers for context-aware ticket resolution. It classifies, drafts, reviews, and escalates support tickets through a multi-step reasoning loop using LangGraph’s dynamic state machine.

🚀 Features

• Ticket Classification: Categorizes incoming support tickets using an LLM.
• Context Retrieval: Uses FAISS + Hugging Face embeddings to retrieve relevant support knowledge.
• LLM Drafting: Automatically drafts a response using Azure OpenAI based on category and context.
• Review & Feedback Loop: Simulates a quality reviewer that can approve or reject drafts with feedback.
• Escalation Path: If a draft is rejected more than twice, the ticket is escalated and logged.

📁 Project Structure

. ├── main.py # Graph definition and execution ├── nodes/ │ ├── classify_node.py # Classifies ticket category │ ├── retrieve_node.py # Retrieves relevant context using FAISS │ ├── draft_node.py # Drafts response using LLM │ ├── review_node.py # Simulates LLM review of draft │ └── escalate_node.py # Escalates and logs tickets after 2 failed reviews ├── utils/ │ └── rag_utils.py # FAISS setup with HuggingFace embeddings ├── .env # Azure OpenAI credentials ├── escalation_log.csv # Log of escalated tickets └── README.md

🧠 Technologies Used

• 🧩 LangGraph: To model the reasoning flow between nodes.
• 🤖 Azure OpenAI: For classification, drafting, and review.
• 🔍 FAISS + Hugging Face: For semantic search using all-MiniLM-L6-v2.
• 🧪 LangChain: For LLM chaining and prompt management.

⚙️ Setup

1. Clone this repo:

bash git clone https://github.com/your-org/support-ticket-agent.git cd support-ticket-agent

1. Create and activate Python 3.10 virtual environment:

bash py -3.10 -m venv venv venv\Scripts\activate # Windows

1. Install dependencies:

bash pip install -r requirements.txt

1. Add your .env file:

OPENAI_API_KEY= OPENAI_API_BASE= OPENAI_API_VERSION= OPENAI_API_TYPE= OPENAI_DEPLOYMENT_NAME=

1. Run the app:

bash python main.py

🧪 Example

python ticket = { "subject": "Charged twice for subscription", "description": "I was billed twice this month for the same plan." }

Depending on review outcomes, the agent will loop through retrieve → draft → review. If rejected twice, it logs the issue via escalate_node.

📌 Notes

• FAISS is initialized once in rag_utils.py using Hugging Face sentence transformers.
• You can expand the knowledge base by editing the docs_by_category dictionary.
• Escalated tickets are saved to escalation_log.csv.