AI Full-Stack Solution Based on Domestic Technology for Patent Technical Document Classification and Summarization

L&F has been reviewing the introduction of an IP service that can efficiently search, classify, and summarize global patent data to drive innovation in secondary battery cathode material R&D. In particular, to protect technical information that is a core asset of the company, building an on-premises environment with enhanced security—rather than an external cloud—was essential.

To this end, Wert Intelligence collaborated with NPU specialist Rebellions to build an NPU-based AI full-stack solution combining the domestic AI semiconductor ATOM™ with the AI model PlutoLM. This solution was designed to process over 300 million pieces of global patent data.

The solution provides two core features:

• AI Classification Copilot: Once a user labels a minimum of 10 documents, it automatically classifies hundreds to thousands of patents within seconds, accurately reflecting user intent through chatbot-style guidance.
• AI Summarization Copilot: It recognizes and extracts key content from patent documents, such as technical objectives, technical solutions, and novelty.

L&F secured security by deploying this on-premises within its internal server architecture. Test results showed a 2.8x improvement in power efficiency and a 16% increase in inference speed compared to existing GPUs.