SupportTicketCategorization

Prototype: React + TypeScript frontend + Flask Python backend that categorizes support tickets into frequent categories and assigns simple priority scores. The backend uses a local transformers pipeline (with configurable MODEL_NAME) and includes batching + caching to speed up inference. The frontend loads tickets.json from frontend/public/, sends it to the backend, and shows results with filters & sorting. The file is the initial state for the categorization, but can be altered in the frontend directly via an input field.

---

Table of contents

• Purpose
• Features
• Architecture
• Prerequisites
• Running locally (quickstart)
• Usage

---

Purpose

Help a support manager quickly understand recurring issues in a small batch (e.g., 50) of support tickets by:

• assigning a human-readable category from a fixed set,
• assigning a simple priority level (Low / Medium / High),
• providing a minimal web UI to inspect, filter, sort and explore results.

This repo is a prototype — intended to be simple to run locally and easy to extend.

---

Features

• Zero-shot classification of tickets into categories (local transformers pipeline).
• Priority detection based on keywords, internal flags and requester frequency.
• Batching and caching to improve throughput.
• React + TypeScript frontend:
  • loads tickets.json from frontend/public/,
  • sends tickets to backend API route /analyze,
  • shows results below input,
  • filters by category and priority,
  • sorting (subject, category, requester, priority).
• Dockerfile for backend & frontend + docker-compose.yml to run everything.

---

Architecture

• backend/ — Flask app exposing /analyze and /health. Loads model once at startup and performs batched classification.
• frontend/ — Vite + React TypeScript app. UI for payload editing, filters, sorting, results table.
• Communication: frontend POST /analyze → backend http://localhost:5000/analyze.

Note: This project does not use the Hugging Face Inference API. Classification is performed locally via transformers.pipeline or could be replaced with a rules-based classifier if you prefer no external model.

---

Prerequisites

• Python 3.10+ (for backend)
• Node.js 16+ and npm (for frontend)
• Docker & docker-compose (optional for containers)
• Enough disk/ram to download + run the chosen transformers model (CPU inference is possible but model weights can be large).

---

Running locally

Docker-compose (project root)

  docker-compose up --build

Usage

1. Start both the backend and frontend as described above.
2. Open the frontend in your browser at http://localhost:3000 or http://localhost:5173.
3. Ensure the backend (http://localhost:5000) is running before analyzing tickets.
4. Place your test data in the frontend/src/tickets.json file. A sample is already provided.
5. In the frontend UI:
   • Use the input/upload area to load tickets.
   • Click Analyze to send tickets to the backend.
   • The categorized and prioritized results will appear in a table below the input.
   • Use the filters (priority & category) above the results to refine which tickets are shown.
   • Sort the table columns (subject, category, priority) to quickly find important tickets.

Example table view:

Subject	Category	Priority
"Payment not going through"	Payment Issues	High
"Need refund ASAP"	Returns/Exchanges	High
"Delivery is late"	Shipping Issues	Medium

---