GFLOPS Co., Ltd. and the RIKEN Center for Computational Science (“R-CCS”) announce the publication of a report summarizing the deployment and operation of the generative AI chat “AskDona※1” on the user support site (the “Fugaku support site”) for the supercomputer Fugaku.
Reading the report
The full report is available at: https://askdona.com/askdona-riken-report/
Report highlights
AskDona has established itself as users’ “thinking partner”
AskDona handles an average of roughly 680 questions per month and has become established as an everyday support tool. Over the 10 months after deployment, the share of “complex queries※2” that require integrating multiple sources grew about 4.3×, from 2.3% to 10.0%. This suggests users are beginning to see generative AI not as a mere “search tool” but as a “thinking partner” to which they can entrust more advanced problem-solving.
Up to 61% more efficient human support on the Fugaku support site
After consolidating the inquiry channel into AskDona, the number of question tickets※3 sent to human support dropped dramatically, reaching a reduction of up to 61% year-on-year in April 2025. In one case, an inquiry that used to take a staff member about four hours was resolved in about five seconds — a major contribution to fast, around-the-clock support and to shifting support staff toward higher-value work.
Demonstrating outstanding performance with proprietary Agentic RAG technology
We comparatively evaluated answer accuracy for “complex questions” against major cloud RAG (Retrieval-Augmented Generation)※4 services (Azure, GCP, AWS) and an OSS framework (LangChain). While the compared systems averaged 61 points, AskDona scored an average of 83 points — 22 points higher. It showed a marked difference especially in “comprehensiveness” (integrating multiple pieces of information) and specialized “practical usefulness,” making its technical superiority clear.
By publishing the insights from this report, we will further promote the real-world implementation of generative AI technology.
About GFLOPS Co., Ltd.
GFLOPS provides AI solutions that support corporate efficiency and innovation, with a strength in proprietary solutions combining large language models (LLMs) and RAG (Retrieval-Augmented Generation) 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
- Website: https://gflops-ai.com/
About the RIKEN Center for Computational Science
Within RIKEN — Japan’s only comprehensive research institute for the natural sciences — R-CCS aims to serve as a world-class, core research center for next-generation computational science in Japan. Beyond R&D toward the next-generation supercomputer succeeding Fugaku, its results — from collaboration with new computing technologies such as quantum computers to developing foundation models for the rapidly advancing “AI for Science” — are expected to benefit people’s lives, industry, and the economy.
- Director: Satoshi Matsuoka
- Location: Kobe, Hyogo
- Website: https://www.riken.jp/research/labs/r-ccs/index.html
※1 AskDona: A generative AI chat already deployed by RIKEN R-CCS on the Fugaku support site. Using applied RAG technology, it answers user questions while referencing Fugaku’s manuals and technical documents. Site: https://askdona.com
※2 Complex query: A “query” is the message (question) a user sends. A “complex query” is a question that is hard to answer from a single source alone and that requires cross-referencing multiple documents and integrating and analyzing their information from multiple angles to arrive at an appropriate answer.
※3 Question ticket: When a Fugaku user contacts support, an individual inquiry is issued through the inquiry system. Each question or request from a user is managed as one “ticket.”
※4 RAG (Retrieval-Augmented Generation): Technology that has a large language model reference external documents and data sources in real time when generating an answer. It helps reduce the risk of hallucination and improves answer accuracy.