Download PDF

　(Head Office: Chiyoda-ku, Tokyo; CEO: Osamu Hata; hereinafter referred to as WAP) participated in the "AI Challenge Day," a generative AI contest held at the Microsoft AI Co-Innovation Lab Kobe from April 18 to 19. The company participated in the AI Challenge Day, a generative AI contest held at the Microsoft AI Co-Innovation Lab Kobe on April 18-19, and won the "World Wide Learning Award. The award was given in recognition of their efforts in dealing with notational distortion by making full use of Sudachi, an OSS developed in-house, as well as their efforts to put the content of their thesis into practice.

WAP Contestant Members

Challenge transcending national and language barriers Careful pre-processing was highly evaluated

　This event, organized by Kadokawa Ascii Research Institute, Inc. (ASCII) and sponsored by Microsoft Japan Co.

　From WAP, four volunteers with diverse workplaces and languages participated in the event: Bharat Singh Bhandari, who is engaged in system development in Singapore; Takashi Yamamura, who is engaged in R&D at the Tokushima NLP Laboratory; Masato Sasano, who belongs to the Product Management Division in Japan; and Lopez Andrade Angel Anthony, who also works at the Tokushima NLP Laboratory. The four contestants were from a variety of locations and languages. Since some of the team members were not usually involved in the development of generative AI, they worked as a team to catch up on RAG*1, and on the day of the competition, communication within the team was unified in English, with team members translating various explanations as necessary.

　As a result, the team won the "World Wide Learning Award" out of four awards. The team was highly evaluated for its use of the company-developed OSS "Sudachi "*2 to deal with distorted notation, data reading using Azure AI Document Intelligence, and chunking that incorporated the content of the most recent paper into practical application. He was critiqued as a "preprocessing demon. Upon receiving the award, contestant member Yamamura commented, "I learned a great deal. I would like to apply the new knowledge to our products and services," he reflected in his participation report blog.

Click here to read the participation report blog by the contestants.

WAP contestants (left) receiving their awards and Mr. Shinsuke Hanagasaki of Microsoft Japan (right), a member of the judging panel

*1: "Retrieval Augmentation Generation (RAG) is an architecture that extends the functionality of large-scale language models (LLMs) such as ChatGPT by adding information retrieval systems that provide grounding data. Adding an information acquisition system allows you to control the grounding data used by the LLM when creating a response."
(Source: Microsoft, Microsoft Build: https://learn.microsoft.com/ja-jp/azure/search/retrieval-augmented-generation-overview

*2: " Sudachi " is a highly flexible 2: "Sudachi" is a Japanese morphological analyzer with high flexibility and accuracy, and is open source software released free of charge by WAP. It incorporates the latest words into the dictionary required for analysis, and as of May 2024, 2.9 million words, the largest number in Japan, have been registered.SudachiPy, a Python version of Sudachi, was released as OSS in June 2019 and surpassed approximately 20.9 million downloads in May 2024.

About AI Challenge Day

The event, organized by ASCII and sponsored by Microsoft Japan, was held in a hackathon format. Members representing each of Microsoft Japan's 10 partner companies gathered in Kobe to compete for ideas and the use of generative AI technology. The full contest can be viewed on ASCII's YouTube archive.

<Outline>
Date: April 18-19, 2024
Location: Microsoft AI Co-Innovation Lab Kobe (Kobe, Japan)
Purpose: RAG-themed hackathon
Organizer: ASCII
Sponsor: Microsoft Japan

<Project Description>
World Heritage Travel Assistant
Develop an assistant to introduce "World Heritage Sites in Japan"!

RAG (search extension generation) is necessary to have questions about Japan's World Heritage sites answered accurately. In this contest, contestants will compete on the accuracy of RAG using Japanese World Heritage data.

<Host Section>
1. Question and Answer FAQ Assistant
Evaluate how close you can get to the correct answer to a previously given question about World Heritage sites using GPT-4.
2. Multimodal Assistant
An assistant who answers the question in question from photo data of World Heritage sites.

<Archive Distribution>
Program: How Far Can Generative AI Take Us? ASCII x Microsoft Generative AI Contest Broadcast ( https://ascii.jp/elem/000/004/194/4194225/)
0:59:40 - Presentation by WAP representatives
1:10:15 - Judges' comments on the presentations
2:39:30 - Results Presentation and Awards Ceremony
2:41:13 - Judges' Critiques

<Detailed Report>
ASCII's detailed report of the two-day event can be found here:
https://ascii.jp/elem/000/004/199/4199213/

About Works Applications

Works Applications was founded in 1996 as an ERP package vendor in Japan. With innovative solutions such as no-customization and free version upgrades, we have supported the growth of our customers, mainly major Japanese companies. Believing in the potential of each individual, we aim to be a "growth engine" that maximizes the value of companies and individuals, and we will continue our pursuit of turning "work" into "creation" and making "work" fun.

*Microsoft and Azure are either registered trademarks or trademarks of Microsoft Corporation in the United States and/or other countries.
*Company names, product names and service names are trademarks or registered trademarks of their respective companies.
*The information in this release is current as of the date of the announcement, and is subject to change or withdrawal without notice. Please be aware that the information in this release regarding forecasts and future prospects is uncertain and may differ from actual results.