🎫 AI Support Ticket Triage System

An LLM-powered support ticket triage system that automatically classifies customer support tickets, assigns priorities, detects sentiment, and drafts response messages.

The application combines FastAPI, Streamlit, Gemini 2.5 Flash, SQLite, and Docker to demonstrate an end-to-end AI workflow for automating customer support.

🌐 Live Demo

Dashboard: https://ticket-triage-app.onrender.com

API Docs: https://ai-ticket-triage-system.onrender.com/docs

Note

Hosted on Render's free tier. Services sleep after ~15 minutes of inactivity, so the first request may take 30–50 seconds.

---

✨ Features

• 🎯 Automatic ticket classification
• 🚨 Priority prediction (Urgent → Low)
• 😊 Sentiment analysis
• ✍️ AI-generated reply drafts
• ✅ Structured output using Pydantic
• ⚡ Parallel processing with ThreadPoolExecutor
• 💾 SQLite persistence
• 📊 Interactive dashboard with charts
• 📤 CSV export
• 📈 Built-in evaluation script
• 🐳 Docker Compose deployment

---

🛠 Tech Stack

Layer	Technology
Language	Python 3.12
Backend	FastAPI
Frontend	Streamlit
LLM	Gemini 2.5 Flash
Validation	Pydantic
Database	SQLite
Package Manager	uv
Concurrency	ThreadPoolExecutor
Containerization	Docker & Docker Compose

---

🏗 Architecture

Customer Tickets
        │
        ▼
 Streamlit Dashboard
        │
 POST /analyze
        ▼
 FastAPI Backend
        │
 Gemini 2.5 Flash
        │
 Structured JSON
 (Pydantic)
        │
        ▼
 SQLite Database
        │
        ▼
 Priority Queue
 Charts • Filters • CSV

---

📂 Project Structure

.
├── api.py
├── app.py
├── database.py
├── eval.py
├── sample_tickets.txt
├── Dockerfile
├── docker-compose.yml
├── pyproject.toml
├── uv.lock
├── sample.env
├── screenshots/
└── README.md

---

🚀 Getting Started

Clone the repository

git clone https://github.com/21f3001527/ai-support-ticket-triage.git
cd ai-support-ticket-triage

Configure environment

cp sample.env .env

Add your Gemini API key.

GOOGLE_API_KEY=your_api_key

Get a free key:

https://aistudio.google.com

---

🐳 Run with Docker

docker compose up --build

Open:

• Dashboard → http://localhost:8501
• API Docs → http://localhost:8000/docs

Useful commands:

docker compose up -d
docker compose logs -f
docker compose down
docker compose down -v

---

💻 Run Locally

Install dependencies

uv sync

Backend

uv run fastapi dev api.py

Frontend

uv run streamlit run app.py

---

📊 Model Evaluation

Run:

uv run python eval.py

Example results:

Category Accuracy : 100%

Priority Accuracy : 75%

Sentiment Accuracy: 100%

The evaluation script tests the same /analyze endpoint used by the application.

---

📸 Screenshots

Dashboard

Analysis Results

Saved Queue

FastAPI Documentation

---

⚠ Limitations

• SQLite is intended for single-user demos.
• No authentication or authorization.
• Streamlit is a prototype UI.
• Gemini free tier has daily request limits.
• Ambiguous tickets may occasionally be misclassified.
• Render's free tier does not provide persistent storage, so saved tickets reset after redeployment.

---

🔮 Future Improvements

• Team assignment recommendations
• SLA monitoring
• PostgreSQL
• Authentication & user roles
• Email integration
• RAG-powered replies
• Async background processing

---

👨‍💻 Author

Rajeev Kumar

• GitHub: https://github.com/21f3001527
• LinkedIn: https://www.linkedin.com/in/rajeev245/
• Email: rajeeev90767@gmail.com