SmartDesk AI

Confidence-Based Intelligent Ticket Automation with Human-in-the-Loop

---

Problem Statement

Enterprises receive thousands of IT support tickets daily. Most follow repetitive patterns but are still handled manually, leading to:

• Slow resolution times
• High operational cost
• Inefficient use of skilled resources

---

Solution

SmartDesk AI is an intelligent ticket automation system that:

• Classifies incoming tickets using machine learning
• Matches them with historical tickets using similarity
• Computes a confidence score
• Automatically resolves high-confidence tickets
• Routes low-confidence cases to human agents

This ensures efficiency without compromising reliability.

---

Key Features

• Confidence-based decision engine
• Hybrid AI (classification + similarity)
• Human-in-the-loop validation
• Explainability using similar past tickets
• Audit logging for transparency
• Role-based workflow (User / Support Agent)

---

Quick Start

1. Clone / extract the project

cd SmartDesk-AI

2. Create a virtual environment (recommended)

python -m venv venv
source venv/bin/activate        # Mac/Linux
venv\Scripts\activate           # Windows

3. Install dependencies

pip install -r requirements.txt

4. (Optional) Pre-train the classifier

python model/train_model.py

5. Run the app

streamlit run app.py

App runs at: http://localhost:8501

---

Project Structure

SmartDesk-AI/
│
├── app.py                   ← Main Streamlit application
│
├── model/
│   ├── train_model.py       ← TF-IDF + Logistic Regression training
│   ├── similarity.py        ← TF-IDF based similarity engine
│   └── classifier.pkl       ← Generated model
│
├── data/
│   └── tickets.csv          ← Sample dataset
│
├── utils/
│   ├── confidence.py        ← Confidence scoring
│   └── decision.py          ← Decision logic
│
├── requirements.txt
└── README.md

---

How It Works

Confidence Formula

confidence = (0.6 × similarity_score) + (0.4 × classification_probability)

Decision Logic

Confidence	Action
≥ 0.80	Auto-Resolved
< 0.80	Needs Human Review
Critical keywords	Escalated

---

Human-in-the-Loop (HITL)

Low-confidence or high-risk tickets are routed to support agents for validation. Agents can:

• Approve AI resolution
• Reject and escalate

This ensures safe and governed automation.

---

Sample Ticket Categories

• Login Issue
• Network Issue
• Application Error
• Access Issue
• Hardware Issue

---

Tech Stack

• UI: Streamlit
• ML: TF-IDF + Logistic Regression
• Similarity: TF-IDF cosine similarity
• Visualization: Matplotlib
• Data: Pandas, NumPy

---

MVP Note

This prototype is trained on a small synthetic dataset for demonstration purposes. With real enterprise data, the system can achieve significantly higher accuracy and confidence, enabling large-scale automation.

---

Future Improvements

• Integration with real ticketing systems (ServiceNow, Jira)
• Larger real-world datasets
• Transformer-based embeddings
• Email/Slack notifications
• Role-based authentication

---

Demo

(Demo Video)

---

License

For academic and demonstration purposes.