dkumi12/Customer-Support-Classifier-

Internship Capstone - Thrive-Africa # 🎫 PulseDesk AI - Customer Support Ticket Classifier > **Production-grade ML system for automated customer support ticket classification using fine-tuned DistilBERT achieving 94% accuracy** Led end-to-end development and deployment as **Project Captain at Thrive Africa**, managing a team of 40+ interns while personally owning the complete CI/CD pipeline and multi-cloud production infrastructure. ## 🚀 **Try Live Demo on Hugging Face →** --- ## 📋 Table of Contents - Overview - Key Features - Architecture - Quick Start - API Usage - Model Performance - Deployment - Tech Stack - Project Structure - Development - Monitoring & Observability - Author - License --- ## 🎯 Overview PulseDesk AI is an intelligent customer support ticket classification system that automatically categorizes incoming support requests into four key categories: **Account**, **Billing**, **Technical**, and **Other**. Built with production-grade infrastructure, this system demonstrates enterprise-level MLOps practices including automated CI/CD, containerization, and comprehensive monitoring. ### Business Impact - **⚡ 94% Classification Accuracy** - Reliable automated ticket routing - **🚀 **👉 Test the classifier instantly:** Visit the Hugging Face demo to classify support tickets in real-time! --- ## 🔥 Key Features ### Machine Learning - **Fine-tuned DistilBERT** for efficient text classification - **Multi-class classification** with confidence scores for all categories - **CPU-optimized inference** for cost-effective deployment - **MLflow experiment tracking** for model versioning and metrics ### Backend Infrastructure - **FastAPI** framework with async request handling - **RESTful API** with automatic OpenAPI documentation - **CORS-enabled** for cross-origin frontend integration - **Health check endpoints** for load balancer integration ### DevOps & Monitoring - **GitHub Actions CI/CD** pipeline for automated deployments to AWS and Hugging Face - **Docker containerization** f …