Customer_Support_AI_System

# Customer Support AI System: Automated Ticket Triage & Resolution

### High-Level Architecture, Multi-Task Machine Learning, RAG Knowledge Retrieval & REST API Serving

**Course / Track:** AI & Machine Learning Foundations — Capstone Project

**Author:** Machine Learning Engineer

**Domain Dataset:** SaaS-Tech Support Operations (`SaaS_Tech_data.csv`) & Knowledge Base (`faq_saas.csv`)

**Core Capabilities:** Automated Categorization, Priority Scoring, Sentiment Analysis, FAQ RAG Matching, and FastAPI Production Serving

---

## Executive Summary & Business Context

Fast-growing digital platforms process hundreds to thousands of customer service tickets daily through diverse channels (web portals, mobile apps, email helpdesks, and live chat). In traditional customer support workflows:

* **Manual Review & Triage Bottleneck:** Support agents spend 30–40% of their shift manually reviewing ticket text, determining department routing, and tagging urgency. This inflates First Response Time (FRT) and creates ticket backlogs.

* **Support Agent Burnout:** Agents repeatedly re-type answers to recurring questions (FAQs) rather than addressing complex technical issues.

* **Unintelligent FIFO Queues:** High-value enterprise customers and urgent production outages sit in standard First-In, First-Out queues alongside routine requests.

* **Customer Churn Risk:** Angry or frustrated customers whose issues are delayed face high churn probability due to lack of immediate emotion-aware escalation.

This project designs and implements an enterprise-grade, end-to-end **Customer Support AI System** that automates ticket categorization, prioritizes issues by severity, analyzes customer sentiment, and retrieves grounded answers from a verified Knowledge Base (RAG).

The notebook is divided into three comprehensive parts:

1. **Part 1: System Analysis & Design (High-Level Architecture):** Clear definition of inputs/outputs, modular system components, end-to-end data flow, AI touchpoints, REST API role, and production technology stack.

2. **Part 2: AI System Lifecycle (Conceptual Framework):** Data collection protocols, PII masking, specialized model inventory, training & cross-validation strategy, and operational business rules.

3. **Part 3: Implementation, Benchmarking & Testing:** Exploratory Data Analysis, text preprocessing, multi-model training & benchmarking (Category, Priority, Sentiment), RAG-based FAQ semantic retrieval, unified pipeline orchestration, production FastAPI deployment testing, and interactive Streamlit UI code.