GFLOPS Co., Ltd. (“GFLOPS”), JAFCO Group Co., Ltd. (“JAFCO”), and RIKEN announce that GFLOPS’s generative AI assistant “AskDona※1” has been deployed on the support site for the supercomputer Fugaku (the “Fugaku support site”) to answer inquiries about using Fugaku quickly and accurately.
Photo: from left, Shoji (Director, Operations and Computer Technologies Division, RIKEN), Mikami (Senior Technical Staff), Nakamura (Engineer), Morimoto (CEO, GFLOPS), and Suzuki (Co-Representative)
Background
The Fugaku support site provides technical support to Fugaku users in Japan and abroad. While comprehensive technical information is important, the sheer volume — combined with the many technical documents created in various formats — can make it hard for users to find what they need and solve their problems. Promoting self-resolution through a powerful search service and a high-accuracy automated answering service had become a pressing challenge for the support site.
Against this backdrop, as part of JAFCO’s startup support program “First Leap※2,” RIKEN and GFLOPS exchanged technical views on generative AI, which sparked discussions about adopting AskDona. After roughly six months of validation and evaluation, the AskDona solution was highly rated and adoption was decided.
Purpose
As part of enriching the Fugaku support site’s FAQ pages and more, this brings a real-time service that delivers answers to users in a chat format, harnessing the effectiveness of question-answering systems built on rapidly advancing generative AI. By training the Retrieval-Augmented Generation (RAG)※3 that underpins AskDona on Fugaku’s manuals and other materials, the answers are optimized to each Fugaku user’s needs.
Why AskDona was chosen
1) Answer accuracy
It cleared the review criterion of 80%+ answer accuracy using RAG technology (patent pending※4). We built a RAG that draws answers not only from Fugaku’s PDF manuals of over 16,000 pages but also from HTML user guides and FAQ pages. Using its patent-pending mechanism for achieving high answer accuracy from large volumes of data, GFLOPS met RIKEN’s technical review requirements for answer accuracy.
2) Usability
Applied RAG technology provides a solution that includes the path to knowledge search. Here, in addition to an interactive chat format powered by RAG, a document search feature is implemented. The document search feature uses the RAG mechanism but does not generate answer text; instead it provides a list of highly relevant documents. By integrating AskDona with the Fugaku support site’s existing ticket-creation feature※5, AskDona seamlessly links the history to a Fugaku support-site ticket when human handling is needed — greatly improving user convenience.
Comments
Fumiyoshi Shoji, Director, Operations and Computer Technologies Division, RIKEN Center for Computational Science
Just as we were thinking generative AI might dramatically improve user support for Fugaku, JAFCO introduced us to GFLOPS and we began exploring it. We quickly saw that GFLOPS’s AI assistant AskDona was more effective than we had imagined — but even more striking was the speed of their response to our requests. We hope to keep collaborating to further enrich services for Fugaku users.
Maria Morimoto, CEO, GFLOPS Co., Ltd.
We began researching RAG as a technology that makes generative AI easy for businesses to use from our founding, and with customer-development support drawing on JAFCO’s broad network, it is a great honor to have RIKEN use AskDona this time. The opportunity to have AskDona used — with feedback from RIKEN — on the Fugaku support site used by people around the world is extremely valuable, and we are confident it will greatly contribute to promoting practical use of applied generative AI technology and to improving user satisfaction.
About AskDona
AskDona is a generative AI assistant that handles SMS, social media, web, and business-chat conversations in one place. It learns internal knowledge and generates answers automatically. It also integrates more than ten task-execution tools that can connect with business applications, autonomously carrying out work based on instructions. At its core, AskDona uses a RAG (Retrieval-Augmented Generation) system built on cutting-edge large language models such as ChatGPT, Claude, and Gemini. Its high-accuracy RAG technology and its mechanism for connecting to RAG from multiple channels simultaneously are patent pending. Site: https://askdona.com
About GFLOPS Co., Ltd.
Founded by two former Google employees supported by JAFCO Group’s entrepreneur program “First Leap.” Aiming for hybrid work between AI and people, it provides solutions leveraging applied generative AI technology.
- Company: GFLOPS Co., Ltd.
- Representatives: Maria Morimoto (CEO); Ryosuke Suzuki (Co-Representative)
- Head office: Shibuya-ku, Tokyo
- Business: Development and provision of AI services leveraging large language models (LLMs) and generative AI technology
- Web: https://gflops-ai.com/
About JAFCO Group Co., Ltd.
Since its founding in 1973, together with its stakeholders, JAFCO has used new technology to tackle challenges beyond existing constraints and conventions. Under the motto “Investing in challenge to create a cycle of growth,” it is committed to creating innovative technologies and services around the world.
- Company: JAFCO Group Co., Ltd.
- President: Keisuke Miyoshi
- Head office: 24F Toranomon Hills Mori Tower, 1-23-1 Toranomon, Minato-ku, Tokyo
- Web: https://www.jafco.co.jp/
About RIKEN
As Japan’s only comprehensive research institute for the natural sciences, RIKEN advances research across a wide range of fields including physics, engineering, chemistry, mathematical and information sciences, computational science, biology, and medical science. To bring research results to society, it conducts joint and commissioned research with universities and companies and actively promotes technology transfer of intellectual property to industry.
- President: Makoto Gonokami
- Location: 2-1 Hirosawa, Wako, Saitama
- Web: https://www.riken.jp/
※1 AskDona: A generative AI assistant that handles SMS, social media, web, and business-chat conversations in one place. It learns internal knowledge and generates answers automatically, and integrates more than ten task-execution tools that can connect with business applications to autonomously execute tasks based on instructions. At its core it uses a RAG (Retrieval-Augmented Generation) system built on cutting-edge large language models such as ChatGPT, Claude, and Gemini.
※2 First Leap: The name of a program run by JAFCO — the venture capital firm with the largest cumulative number of investments in Japan — to support entrepreneurs.
※3 Retrieval-Augmented Generation (RAG): Technology for giving large language models external information, combining information retrieval and generation.
※4 Patent pending: A patent is pending for technology that improves the answer accuracy of Retrieval-Augmented Generation (RAG).
※5 Ticket-creation feature: A feature whereby a Fugaku user submits an inquiry via a form, a ticket is created, and an inquiry handler can respond.