Patentfield Wins Special 'Technical Novelty Award' in Generative AI Development for Government Examination Processes at METI/NEDO's 'GENIAC-PRIZE'

Patentfield Inc. (Headquarters: Nakagyo-ku, Kyoto; Co-CEOs: Naoya Murakami, Kosuke Ishizu) has received the special 'Technical Novelty Award' in Area 02 'Development of Generative AI Contributing to the Efficiency of Examination Operations in Government Agencies' of the NEDO Prize Money Application Program 'GENIAC-PRIZE', sponsored by the Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO). About 'GENIAC-PRIZE' 'GENIAC-PRIZE' is part of the 'GENIAC (Generative AI Accelerator Challenge)' project promoted by METI and NEDO to strengthen the development capabilities of generative AI foundation models in Japan. It utilizes the NEDO Challenge program system to evaluate the outcomes of generative AI applications developed and demonstrated based on themes expected to be solved by generative AI services. In Area 02, focusing on improving the efficiency of examination tasks—a core operation of government agencies—the development of a prototype using 'patent examination operations' as a model case was required. Comprehensive evaluation was conducted through practical operation tests by patent examiners (performance in information retrieval, matching location display, rationale output, and usability) and document screening of proposals. *For details on the GENIAC-PRIZE, please refer to the official website. https://geniac-prize.nedo.go.jp/ About the Special 'Technical Novelty Award' The 'Technical Novelty Award' is given to entries that have undertaken advanced technical challenges, such as AI agent orchestration and novel architectures. Overview and Technical Features of the Award-Winning System The 'Patent Prior Art Search Support AI System Reproducing the Examiner's Cognitive Process' developed by our company is a system that uses AI to replicate the cognitive process of examiners: 'Claim structuring -> Evidence search -> Comparative judgment'. Built on a proprietary patent-specialized foundation model continuously pre-trained on over 100 billion tokens of expert data, such as patent gazettes and examination guidelines, it consists of three core AIs: claim structuring, cited paragraph matching, and citation judgment/rationale generation. Major technical features include 'promptless inference' without instruction texts via Supervised Fine-Tuning (SFT), a significant improvement in Recall@100 (from 42.22% to 70.22%) through contrastive learning using examiners' citation records as training data, and a fully local AI architecture that uses no external APIs, ensuring high security and sustainability. Going forward, we will consider deploying the technologies cultivated in this system to our 'AI Patent Comprehensive Search and Analysis Platform Patentfield'. Award Comment We are deeply honored to receive the Technical Novelty Award (Special Award). Patent examination is a highly specialized task that requires identifying essential technical commonalities and differences from a vast amount of literature. We are very pleased to have taken on the challenge of reproducing this cognitive process with AI and to have demonstrated a fraction of the effectiveness of our expert AI model group uniquely trained on patent data. We will continue to dedicate ourselves to the development and social implementation of AI technology specialized for patent operations, aiming to contribute to the construction of a sustainable and secure 'patent search and examination infrastructure'. About Patentfield Inc. Provides the 'AI Patent Comprehensive Search and Analysis Platform Patentfield', which structures and summarizes over 80 million patent records using generative AI, improving the efficiency of expert search operations by up to 80%. Under the slogan 'From reading patents to seeing them', we strongly support intellectual property strategies and R&D. https://patentfield.com/ Patentfield Inc. Corporate Team Email: info@patentfield.com