Purpose : Our goal is to propose an efficient and scalable Q&A bot platform. This platform is designed to enhance LLM performance and streamline chatbot construction by integrating various hight-quality libraries and development tools. Additionally, we aim to create an application that allows for immediate correction of chatbot responses if they are incorrect.
<br>Methods :We apply concepts such as Multi LLM, multi RAG, and multi embedding. Additionally, we integrate a similarity-based Chatbot by invoking the similarity document retrieval module in the RAG-based LLM platform.
<br>Results : We enabled the RAG-based LLM to generate responses by first uploading documents without scenario registration. Additionally, to address cases where LLM couldn't respond, we registered query-response sets in the intent, facilitating easy correction of Chatbot responses.
<br>Conclusion : We propose a flexible Chatbot architecture. Current research lacks prompt responses to LLM errors, and traditional Chatbots need retraining for intent changes. We suggest integrating a learning-free Chatbot for instant error conrrection during chats, with reporting and rectifying measures.
