Asian Journal of Research in Computer Science

This paper presents the design and implementation of a real-time, embeddable customer service system that integrates Retrieval-Augmented Generation (RAG) with human-in-the-loop (HITL) escalation. The system addresses common limitations in traditional customer support, such as slow response times and inconsistent service, by combining intelligent automation with timely human intervention. Built using FastAPI, ReactJS, PostgreSQL, and Centrifugo for WebSocket communication, the architecture supports low-latency interaction and seamless transitions between bot and human agents. At its core, the system leverages OpenAI’s GPT-3.5-turbo model to generate responses informed by user-provided domain-specific intent files.

To evaluate the effectiveness of the system, we conducted controlled testing using metrics such as average response time, session resolution rate, escalation latency, and accuracy of chatbot replies. Results showed that 71% of user queries were successfully handled by the chatbot alone, with an average response time under 3 seconds and human agent intervention occurring in less than 3 minutes for escalated cases. These outcomes suggest that the proposed system offers a scalable and intelligent alternative for real-time customer engagement, particularly in environments where domain-specific support and rapid escalation are essential.