Large-scale language model system

The system addresses inefficiencies in RAG databases by constructing field-specific RAG databases with image integration and feature vectorization, enhancing storage and reducing costs, thus providing accurate and responsive large-scale language model systems.

JP2026111060AActive Publication Date: 2026-07-03INST OF MEDICAL INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
INST OF MEDICAL INFORMATION TECH CO LTD
Filing Date
2024-12-23
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Large-scale language models face challenges in efficiently incorporating up-to-date knowledge, handling image data, and preventing hallucinations, especially in domains like medicine, due to high computational costs, data redundancy, and inefficient search methods in Retrieval-Augmented Generation (RAG) databases.

Method used

A system that constructs multiple RAG databases for specific fields, extracts relevant feature vectors, removes duplicates, and integrates image data, enabling efficient search and retrieval of contextual and background information without retraining the model, using a page image acquisition, text extraction, and feature vectorization process.

Benefits of technology

Enhances storage capacity, reduces computational costs, and prevents hallucinations by providing accurate and comprehensive answers, including image data, while maintaining responsiveness to urgent information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2026111060000001_ABST
    Figure 2026111060000001_ABST
Patent Text Reader

Abstract

As needed, multiple Search-Augmented Generative (RAG) databases are constructed for each related field to increase the storage capacity for additional information. Based on the chunks obtained from RAG searches, the entire text or image of the original document pages containing relevant content is referenced to eliminate missed references. At the same time, background information and related information that may be present around the searched chunks are reflected in the context of the question, and the records of the question and answer are compiled into a searchable database, providing a large-scale language model using search-aggregated generative generation that eliminates the need for costly and time-consuming re-generating of answers. [Solution] The RAG database includes means for acquiring page image data, means for recording page text in a database, means for extracting related feature vectors, and means for recording feature vectors in a database.
Need to check novelty before this filing date? Find Prior Art