Sharvandeep/ticket-routing-openenv

🚀 Ticket Routing OpenEnv Environment

🌍 Overview

This project simulates a real-world customer support ticket routing system, where an AI agent classifies user complaints into the correct department, assigns priority, and decides escalation.

It is built as an OpenEnv-compatible environment to evaluate intelligent agents on realistic enterprise workflows.

🎯 Motivation

Customer support systems handle thousands of tickets daily. Incorrect routing causes delays and poor user experience.

This environment models that challenge and helps:

• —Train AI agents
• —Evaluate decision-making
• —Benchmark automation systems

⚙️ Core Features

• —Real-world task simulation
• —Multi-step environment
• —3 difficulty levels (easy, medium, hard)
• —Deterministic grading (0.0–1.0)
• —Reward-based learning system
• —REST API endpoints
• —Hugging Face deployment

🧠 Task Design

Each ticket must be classified into:

• —Department → billing / technical / account / general
• —Priority → low / medium / high
• —Escalation → yes / no

Difficulty levels:

• —Easy → clear tickets
• —Medium → moderate ambiguity
• —Hard → complex interpretation

🧮 Reward Function

Reward is calculated per step:

• —Correct department → +0.4
• —Correct priority → +0.3
• —Correct escalation → +0.3

Total reward range: 0.0 to 1.0

This ensures:

• —Partial learning signals
• —Continuous feedback
• —Realistic evaluation

🔌 API Endpoints

📥 Example

Reset: POST /reset

Step: { "department": "technical", "priority": "high", "escalation": "yes" }

🤖 Baseline Agent

A baseline agent is provided in inference.py.

It uses a hybrid approach:

• —Rule-based logic (stable)
• —Optional LLM (adaptive)

This ensures:

• —Reliability
• —Reproducibility
• —Compatibility with evaluation

🧪 Setup (Local)

pip install -r requirements.txt uvicorn app.main:app --reload

🐳 Docker

docker build -t ticket-env . docker run -p 7860:7860 ticket-env

Agent

The inference script uses an LLM (via OpenAI-compatible API) with a fallback rule-based system to ensure stability and consistent performance.

📊 Evaluation

• —Works with OpenEnv validation
• —Deterministic grading
• —Runs within constraints

💡 Highlights

• —Real-world use case
• —Clean design
• —Strong reward system
• —Fully deployable
• —Agent evaluation ready

🏁 Conclusion

This environment provides a practical benchmark for AI agents in customer support automation, combining real-world relevance with structured evaluation.