Design and Construction of QA (Question Answering) System Based on Artificial Intelligence Using FAISS and Mixtral Language Model

Abstract

The development of technology and the increasing amount of digital data drives the need for a fast, accurate, and easy-to-use information retrieval system. This study aims to build a Question Answering (QA) system or question answering system based on artificial intelligence. This system is designed to be able to answer user questions based on the contents of text documents, especially PDF documents. In this study, two main technologies are used, namely FAISS and Mixtral Language Model. FAISS is used to search for the most relevant text fragments from documents based on similarity of meaning (vector similarity), while Mixtral is used to understand questions and produce answers in the form of natural sentences. The steps for building the system start from collecting document data, text pre-processing, converting text to vectors, building an index using FAISS, to developing an answering module using Mixtral. The test results show that the system is able to provide answers that are in accordance with the context, with a fairly fast processing time. The advantage (novelty) of this study lies in the combination of FAISS and Mixtral which are applied modularly and specifically developed to answer questions based on local documents. This has not been widely applied in similar QA systems, especially in Indonesia. This system has great potential for use in the fields of education, digital libraries, legal services, and document management in companies.