By utilizing Enterprise Search QuickSolution® provided by Sumitomo Electric Information Systems Co., Ltd., Sumitomo Electric Industries, Ltd. built a large-capacity RAG infrastructure on a group-wide scale in the short space of about two weeks. This infrastructure uses RAG technology to search and extract necessary information and link it to generation AI, providing a function for highly accurate question answering from hundreds of TB of internal information stored in file servers, etc. As a result, approximately 290,000 people in the Sumitomo Electric Group can now collect internal information in an interactive format, enabling improved productivity and digital transformation through the effective use of knowledge.
With the emergence of ChatGPT in 2022, many companies have come to recognize the importance of using generative AI and are exploring the best ways to use it. Sumitomo Electric is also actively promoting the use of generative AI, and in October 2023, it rolled out an environment in which ChatGPT can be safely used to all group companies. As the use of this environment progressed, many users in each department expressed their desire to use internal information with generative AI, highlighting the need for a generative AI that can respond to internal information.
The key to solving the problem was QuickSolution ® and RAG*1. QuickSolution ® is a domestically developed enterprise search that can search the contents of files across file servers and various internal systems. QuickSolution ® utilizes RAG to provide a generative AI integration function that answers questions based on the results of searching internal information.
Sumitomo Electric uses QuickSolution® as a group-wide search platform, and has used the generation AI linkage function of QuickSolution® to build a RAG platform capable of answering internal information. This makes it possible to extract necessary information from file servers (400TB) and document management systems (1.8 million documents) across systems, and generate answers based on that content. It is noteworthy that the construction of a large-capacity RAG platform of several hundred TB was completed in just two weeks. This was possible because, while a normal RAG system requires data to be uploaded to the cloud to be linked to the generation AI, the generation AI linkage of QuickSolution® does not require changing the storage location of the data.
This RAG platform has three features. The first is that it provides accurate answers. The quality of RAG answers depends on the accuracy of the extracted documents, but this platform provides accurate answers by linking reliable internal information narrowed down by QuickSolution® ‘s high-precision search to the generation AI. The second is that it can take into account the user’s viewing permissions. QuickSolution® can search and extract documents across the board based on access rights, so it can provide answers based on appropriate permissions from hundreds of TB of internal information distributed across multiple systems. The third is that it is secure to operate. It is designed*2 not to allow the linked generation AI to learn the content of the questions, so there is no risk of questions containing personal or confidential information being leaked.
Currently, the RAG platform is available to anyone who applies – approximately 40,000 people in the Sumitomo Electric Group in Japan, and approximately 290,000 people overseas – and is reaping benefits such as improved productivity and the promotion of digital transformation through the effective use of knowledge. Below are some examples.
Sumitomo Electric plans to continue to actively promote the use of generative AI. It plans to continuously improve its RAG platform and also plans to expand the data sources that serve as the platform’s information sources. The company aims to create a competitive organization by making the most of the RAG platform realized through the integration of QuickSolution® search and generative AI, and by promoting the efficiency of information searches and the effective use of internal knowledge.
SOURCE: PRTimes
