SCALABLE AND REAL TIME EMBEDDABLE RETRIEVAL-AUGMENTED GENERATION (RAG) ANALYTICS SYSTEM FOR CUSTOMER SERVICE

Abstract

Customer service has transitioned from traditional face-to-face and phone-based interactions to digital platforms that emphasize speed, scalability, and personalization. Despite these advances, AI-driven tools like chatbots often face challenges in contextual understanding, handling multi-step queries, and enabling smooth escalation, which can lead to dissatisfaction. This research develops a scalable, real-time embeddable Retrieval-Augmented Generation (RAG) analytics system that integrates AI efficiency with human adaptability. The system architecture employs FastAPI, Celery, and Centrifugo for backend processing, ReactJS with Vite for the frontend, and PostgreSQL for secure data handling. It incorporates OpenAI’s GPT-3.5-turbo API for natural language processing and NovuHQ for real-time notifications, ensuring context-aware responses and timely human intervention. An iterative development model guided the design, enabling incremental refinements through continuous feedback from customers, agents, and administrators. Key features include iframe embedding, direct web links, reusable components, real-time chat, Google OAuth authentication, session tracking, analytics, and escalation pathways. Testing confirmed that the system effectively handles routine queries while seamlessly escalating complex cases to human agents. Evaluation results highlight improved scalability, reduced response time, and preserved personalization. Its embeddable design supports adoption across diverse sectors, including SMEs and educational institutions. Future extensions will explore multilingual capabilities, sentiment-driven escalation, and CRM integration for holistic customer relationship management.