Rancang Bangun Chatbot Customer Service Multi-Tenant Berbasis Large Language Model Menggunakan Context Stuffing

Abstract

The advancement of Artificial Intelligence (AI), particularly Large Language Models (LLMs), has transformed customer service into a more automated, adaptive, and responsive system. However, LLM-based chatbots still face several challenges, including hallucination, limited capability to serve multiple organizations simultaneously, and difficulties in website integration. This study aims to design and develop an embeddable multi-tenant customer service chatbot based on a Large Language Model. The system was developed using the Go programming language, Fiber framework, PostgreSQL database, and a JavaScript widget that can be embedded into company websites using a single line of code. The Research and Development (R&D) approach was adopted, while the context stuffing method ensured that chatbot responses were generated only from each tenant's official knowledge base, thereby minimizing hallucination. The evaluation included black-box testing, response accuracy measurement, hallucination assessment, response time analysis, and tenant data isolation testing. The experimental results demonstrated that the chatbot achieved 100% response accuracy for in-domain questions, produced zero hallucination for out-of-domain questions, maintained an average response time of approximately four seconds, and successfully preserved complete data isolation among tenants. Furthermore, the chatbot widget was easily embedded into company websites with minimal configuration. These findings indicate that the proposed chatbot provides an effective, secure, and scalable AI-based customer service solution suitable for multiple organizations.