Mindware Research Institute – Long Before Generative AI, We Were Mapping Knowledge

Implementation of an AI system based on the concept formation approach

Have you implemented RAG or an internal AI chatbot, only to find that it fails to retrieve the right information or provide clear evidence for its answers? The quality of AI-generated answers depends not only on the underlying model, but also on how source documents are structured and retrieved. However, rebuilding an entire document collection for AI use can be a major undertaking. Mindware Research Institute creates “LLM Wikis” that connect concise document overviews with precise references to the original sources. They are provided in portable formats that can be integrated into existing RAG systems and AI chat applications, helping AI locate the right documents and generate answers grounded in the original content.

It is also an interface between humans and AI.

Mindware is a vision that our founder, Kunihiro Tada, has pursued since the 1990s. It refers to products and services that help people achieve, by themselves, results close to those produced by superior intelligence or talent. Since 2000, we have accumulated technology and experience in fields such as self-organizing maps, probabilistic reasoning, statistical analysis, and machine learning. Today, with the rapid progress of LLMs, we are beginning to see a concrete path toward realizing Mindware by combining these accumulated technologies with large language models. The knowledge base technologies provided by Mindware Research Institute are the first step in this direction. They are not intended merely for information retrieval, but as an intellectual foundation that helps people and organizations think, judge, and make decisions. In the future, we plan to develop these technologies into a Digital Advisor / Digital Consultant that continuously accumulates organizational knowledge, experience, and judgment, and supports decision-making in both management and field operations.

Concept Index

ConceptMiner Concept Index adds similarity search capabilities to existing relational databases such as PostgreSQL.It can be used to extract semantically similar groups from data such as VoC records, inquiry histories, open-ended survey responses, sales notes, and case summaries.<br />

ThinkNavi

This platform offers a variety of applications using conceptual structure models. It can be used as a viewer for Mindware content, explore new concepts from data obtained through automated research, and easily build an AI concierge. Furthermore, the AI guides you toward higher-quality thinking through chat.

ConceptMiner

Concept Modeling Engine. A set of libraries with tools, such as Growing Neural Gas (GNG) + Minimum Spanning Tree (MST), Self-Organizing Maps (SOM). Users can easily call these functions from their own systems using APIs, and develop applications using text/data mining, AI explainability, and associative memory.