Retrieva to Exhibit LLM Vulnerability Diagnostic Web App 'KOKOROE' at 'GENIAC-PRIZE Achievement Caravan'

Retrieva, Inc. (Toshima-ku, Tokyo; Takuya Taguchi, Representative Director) is pleased to announce that it will exhibit its LLM vulnerability diagnostic web application, 'KOKOROE', at the 'GENIAC-PRIZE Achievement Caravan,' a program utilizing NEDO grants, to be held at Shibuya Stream Hall in Shibuya-ku, Tokyo, on April 14, 2026. We will be showcasing 'KOKOROE,' which was jointly developed with our group company, Manyo, Inc. 'KOKOROE' is a web application that allows users to diagnose whether an LLM is providing safe responses and to improve the LLM based on the diagnostic results. By simply providing human-readable, straightforward policies, users can tune LLMs to be used safely according to their specific purposes. In recent years, while the adoption of generative AI has progressed in many companies due to expectations for business improvement and productivity, LLM responses are not always safe. There are inherent risks specific to each industry, such as responses that promote criminal activity, dangerous work procedures in manufacturing, or incorrect medical judgments in the pharmaceutical industry. 'KOKOROE' addresses these risks and vulnerabilities by automatically performing diagnostics and extensions using simple policies written by humans. Our company is engaged in research and development that emphasizes 'safety' in addition to 'utility' so that companies can use Large Language Models (LLMs) safely and effectively in their operations. Against this background, we have participated in the 'Development of Risk Exploration and Risk Reduction Technologies for Ensuring the Safety of Generative AI' under the GENIAC PRIZE, promoting the visualization of potential risks inherent in generative AI and the development of technologies to reduce them. At this caravan, we aim to share these achievements widely to promote discussion on improving the safety of generative AI and contribute to overall societal safety. 'GENIAC-PRIZE' is a prize-based contest led by the Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO), aimed at accelerating the social implementation of generative AI. It sets four themes where solutions by generative AI are desired and recruits/judges applications based on specific needs. The four themes are: 1. Formalizing tacit knowledge in the manufacturing industry, 2. Improving productivity in customer support, 3. Streamlining administrative tasks in government agencies, and 4. Developing risk exploration and risk reduction technologies for ensuring the safety of generative AI. The achievement caravan will provide opportunities for networking and exchange with participants, in addition to introducing use cases of generative AI by winners and applicants.