UtkarshTheAIMaster/ai-ticket-routing-pro

🎫 AI Ticket Routing Pro (OpenEnv Environment)

An advanced OpenEnv-compatible Reinforcement Learning environment for simulating real-world customer support systems.

This project challenges an AI agent to intelligently:

• —Classify support tickets
• —Assign appropriate priority
• —Decide escalation
• —Generate professional responses

🚀 Why This Project?

Customer support automation is a real-world, high-impact problem used by:

• —SaaS companies
• —E-commerce platforms
• —Enterprise support systems

This environment replicates real-world complexity including:

• —Multi-step decision making
• —Ambiguous user intent
• —Escalation handling
• —Human-like response generation

🧠 Core Features

🔹 OpenEnv Compatible

Fully follows OpenEnv standards:

• —/reset, /step, /state, /tasks, /grader, /baseline

🔹 Multi-Decision Action Space

The agent must output:

🔹 Rich Observation Space

The agent receives:

• —Ticket message
• —Conversation history
• —Difficulty level (easy/medium/hard)
• —Feedback from previous attempt
• —Reward signals
• —Hints after failures

🔹 Realistic Task Design

• —15 curated tasks
• —Balanced across difficulty:

• —Easy (5)
• —Medium (5)
• —Hard (5)

Includes:

• —Billing issues
• —Technical failures
• —Fraud detection
• —Enterprise outages
• —Legal threats

🔹 Advanced Reward System (0.0 → 1.0)

✔ Partial credit for near-correct answers ✔ Bonus for strong decisions ✔ Feedback-driven learning

🔹 🌐 Interactive Web Dashboard

A modern UI dashboard allows real-time interaction:

• —View tickets
• —Submit agent actions
• —See rewards + feedback instantly

👉 Open in browser:

http://localhost:7860

🏗️ Project Structure

ai-ticket-routing-pro/
│
├── server/
│   ├── app.py              # FastAPI server + endpoints + UI serving
│   └── environment.py      # Core RL logic + tasks + reward system
│
├── models.py               # Pydantic schemas
├── client.py               # Python RL client
│
├── frontend/               # Web dashboard
│   ├── index.html
│   ├── style.css
│   └── script.js
│
├── baseline/
│   └── inference.py        # LLM + rule-based agent
│
├── openenv.yaml            # OpenEnv metadata
├── Dockerfile              # Deployment
├── requirements.txt
└── README.md

⚙️ Setup & Run

1️⃣ Install dependencies

pip install -r requirements.txt

2️⃣ Run server

uvicorn server.app:app --host 0.0.0.0 --port 7860

3️⃣ Open UI

http://localhost:7860

🧪 API Usage

🔹 Reset Environment

POST /reset

🔹 Take Action

POST /step

Example:

{
  "category": "billing",
  "priority": "medium",
  "response": "We sincerely apologize for the inconvenience. Your refund will be processed within 3-5 business days.",
  "escalate": false
}

🤖 Baseline Agent

Includes:

• —Rule-based agent
• —Optional LLM (Groq / OpenAI)

Run:

python baseline/inference.py

📊 Learning Objectives

This environment helps train and evaluate agents on:

• —Decision making under uncertainty
• —Priority reasoning
• —Customer empathy
• —Escalation judgment
• —Real-world workflow simulation

🐳 Deployment (HuggingFace Spaces)

• —SDK: Docker
• —Port: 7860
• —Auto-deploy supported

👤 Author

Utkarsh Meta PyTorch OpenEnv Hackathon x Scaler School of Technology

💡 Final Note

This project goes beyond a simple classification task — it simulates a real-world AI support system with structured decision-making and human-like interaction.