🎫 Ticket Agent

An intelligent multi-role support ticket agent powered by LangGraph and Snowflake Cortex and Streamlit. This system automates the lifecycle of a support ticket—from initial AI-driven triage and priority classification to human-in-the-loop administrative resolution.

✨ Features

• Smart New-Ticket Evaluation – Evaluates new issues (priority + metadata) before ticket creation and response generation
• Snowflake Cortex Search – Retrieves relevant past tickets and knowledge base context to inform responses
• AI + Human Support Loop – Combines AI-generated replies with admin human-in-the-loop intervention in the same chat threads
• Lifecycle Tracking & Resolution – Stores all agent messages and runs resolution checks to keep ticket status up to date

👥 Dual-Role Interface

The system features a unified routing architecture in frontend/app.py that allows seamless switching between two distinct user experiences, managed via a global sidebar toggle.

👤 User Experience

• Support Desk Landing: A dedicated home page to raise new problems or check the status of existing ones.
• Live Status Tracking: Real-time ticket lookups with direct "Open Chat" links to resume active conversations.
• Ticket History: A searchable, tabular overview of all personal past tickets for easy reference.
• AI-Driven Chat: Direct interaction with the Mistral-powered agent for immediate troubleshooting and automated triage.

🔑 Administrator Dashboard

• Queue Management: A comprehensive view of all system tickets with visual color indicators for priority (URGENT to LOW) and status.
• Advanced Triage: Built-in tools to filter the queue by status (OPEN/CLOSED) and sort by urgency, subject, or update recency.
• Human-in-the-Loop Override: Admins can enter any active chat thread to provide manual assistance alongside the AI agent.
• One-Click Resolution: Streamlined controls to close resolved tickets directly from the dashboard or within the chat view.

💬 Hybrid Chat System

• Shared Interface: Both roles utilize the same user_chat.py component, ensuring a consistent history of both AI and human responses.
• Visual Role Cues: The chat interface identifies messages from "Users," "AI Agents," and "Human Agents" (Admins) using distinct avatars and alignments to maintain clarity.

🏗️ Architecture

The agent uses a LangGraph state machine with the following workflow:

stateDiagram-v2
   [*] --> load_history
   load_history --> retrieve

   retrieve --> evaluate_ticket: new_ticket
   evaluate_ticket --> create_ticket
   create_ticket --> generate

   retrieve --> update_ticket: existing_ticket
   update_ticket --> generate

   generate --> store_agent_message
   store_agent_message --> check_for_resolution
   check_for_resolution --> [*]

Loading

Nodes

Node	Description
load_history	Loads past ticket messages
evaluate_ticket	Evaluates ticket priority and metadata for new tickets
retrieve	Fetches relevant context from Snowflake Cortex Search
generate	Generates a support response based on priority and context
create_ticket	Creates a new ticket in the database
update_ticket	Adds a message to an existing ticket
store_agent_message	Persists the agent's response
check_for_resolution	Checks if the issue is resolved and updates ticket status

📁 Project Structure

ticket_agent/
├── backend/
│   ├── agent.py                     # Main agent entry point
│   ├── db/
│   │   ├── snowflake_utils.py       # Snowflake session management utils
│   │   ├── sql_utils.py             # SQL execution helpers
│   │   ├── zen_repo.py              # Ticket repository operations
│   │   ├── csv/                     # CSV files for sample data
│   │   ├── tables/                  # SQL table definitions
│   │   ├── views/                   # SQL view definitions
│   │   └── cortex_search_services/  # SQL cortex definitions
│   ├── graph/
│   │   ├── graph.py                 # LangGraph workflow definition
│   │   ├── router.py                # Conditional routing logic
│   │   ├── state.py                 # State schema definition
│   │   └── nodes/                   # Individual graph nodes
│   └── llm/
│       └── model.py                 # LLM configuration
├── scripts/
│   ├── setup_db.py                  # Database setup script
│   └── chat_with_agent.py           # CLI chat interface
├── frontend/
│   ├── app.py                       # Streamlit app entry point
│   └── views/                       # Streamlit app pages
├── requirements.txt
├── Makefile
└── .env.example

🚀 Getting Started

Prerequisites

• Python 3.10+
• Snowflake account with Cortex enabled
• Mistral API key (used in backend/llm/model.py, you can change the code to use another LLM)
• Environment: Ensure your .env contains your Snowflake credentials and MISTRAL_API_KEY

Installation

1. Clone the repository

   git clone <repository-url>
   cd ticket_agent

2. Create and activate a virtual environment

   make venv
   source venv/bin/activate

3. Configure environment variables

   Copy the example file and fill in your credentials:

   cp .env.example .env

   Required variables:

   Variable	Description
   SNOWFLAKE_ACCOUNT	Your Snowflake account identifier
   SNOWFLAKE_USER	Snowflake username
   SNOWFLAKE_TOKEN	Snowflake password/token
   MISTRAL_API_KEY	API key
   SNOWFLAKE_WAREHOUSE	Compute warehouse name
   SNOWFLAKE_DATABASE	Database containing your data
   SNOWFLAKE_SCHEMA	Schema name
   SNOWFLAKE_CORTEX_SEARCH_SERVICE	Cortex Search service name

Running the Agent

To test the agent through command line, run:

make run_agent
# or
PYTHONPATH=.:backend python scripts/chat_with_agent.py

Running the Frontend

Start the ticketing system UI:

make app
# or
streamlit run frontend/app.py

🛠️ Technologies

• LangGraph – Stateful agent orchestration
• LangChain – LLM integration
• Snowflake – Data storage and processing
• Snowflake Cortex – Semantic search & LLM inference
• Streamlit - App frontend