🎫 TicketTriage: AI-Powered Customer Support Routing

TicketTriage is an intelligent customer support routing system designed to automate the classification and prioritization of inbound tickets. By replacing manual triage with a hybrid architecture, the system instantly analyzes customer messages, categorizes them by department (Accounts, Billing, Orders), and assigns a priority level (Low to Critical).

---

Hybrid Architecture Workflow

graph TD
    Input[Customer Ticket Text] --> Preprocess[Text Preprocessing]
    Preprocess --> Vectorizer[TF-IDF Vectorizer]
    Vectorizer --> ML_Model[Linear SVC Model]
    
    ML_Model -->|Predicts Category| Category[Department: Accounts / Billing / Orders]
    ML_Model -->|Predicts Priority| Priority[Base AI Priority]
    
    Priority --> Rules{Business Rule Engine}
    
    Rules -->|Normal Ticket| AI_Keep[Keep Base Priority]
    Rules -->|Matches 'Fraud', 'Urgent'| Override[Override to CRITICAL]
    
    Category --> Output[(Final Classified Ticket)]
    AI_Keep --> Output
    Override --> Output

Loading

---

Application Dashboard

---

🚀 Key Features

• Hybrid Decision Engine: Combines the predictive power of a Machine Learning model with the absolute safety of a hardcoded Python business rule engine.
• Mutually Exclusive Routing: Ensures that "Critical" emergencies are isolated to a dedicated SWAT queue and completely removed from standard departmental workflows to prevent double-working.
• Real-Time Analysis UI: A clean, interactive Streamlit dashboard for individual ticket testing and triage.
• Bulk Processing Engine: Upload a raw CSV of customer messages and download a consolidated, routed ZIP file of department-specific queues in seconds.

---

🧠 System Architecture

TicketTriage operates on a dual-layer validation system:

1. Layer 1 (The ML Brain): A TfidfVectorizer converts natural language into numerical features, which are then evaluated by a Linear Support Vector Classifier (SVC) to predict the base category and priority.
2. Layer 2 (The Safety Net): Because ML models can misclassify rare or Out-of-Distribution (OOD) data, a hardcoded Business Rule Engine scans the text for severe keywords (e.g., "fraud", "stolen"). If triggered, it forcefully overrides the AI's prediction, escalates the ticket to "Critical", and logs a business rule flag.

---

📂 Project Structure

TicketTriage/
│
├── data/                   # Raw and processed datasets (ignored in git)
├── models/                 # Serialized model (.pkl) files (ignored in git)
├── src/                    # Core pipeline logic
│   ├── preprocessing.py    # Text cleaning and vectorization
│   ├── train_model.py      # ML training script
│   └── predict_bulk.py     # Batch processing and business rule engine
│
├── app.py                  # Streamlit frontend and UI logic
├── requirements.txt        # Python dependencies
└── README.md               # Project documentation