Mindware Research Institute is currently developing a SaaS called ConceptMiner, and we are pleased to announce that we have decided to make a major change to the product concept and focus on the field of AI governance.
For the past 25 years, Mindware Research Institute has been working with a SOM-based data mining system called Viscovery SOMine. Following major advances in LLM, two years ago we began experiments using SOM to create maps of text chunks. Based on these results, in 2025 we replaced the SOM with our own proprietary GNG + MST (growing neural gas + minimum spanning tree) and developed the MVP (Minimum Viable Product) of ConceptMiner. We had completed the system to the point where it was almost ready to be released as a paid SaaS, but after reflecting on whether there was a clear need for it and whether users would recognize the need for it, we decided to continue further development.
LLM’s performance already far exceeds the level required for chat by general users, and enterprise services are expected to begin in 2026. OpenAI already offers “GTP Enterprise” for businesses, with enhanced security and business task automation features, and the company’s CEO, Sam Altman, has begun talking about his vision of an “AI CEO.” He suggests that in the future, AI services may be developed at the level of business management and corporate strategy, going beyond traditional AI chat and task-level automation.
As AI begins to be used in corporate management, the problem of humans being unable to understand “why the AI made that decision” will surface. The LLMs used today are implemented using deep learning neural networks, so their internal workings are a black box. Families such as SOM and GNG, as well as BBN (Bayesian Belief Network), can express AI behavior in explainable models, and will become important interfaces between humans and AI in the future.
Until now, ConceptMiner has been implemented as an analytical tool with an unseparated front-end and back-end, but we have restructured it as a concept network model engine (back-end) so that various applications can be developed simply by developing the front-end. The first such application, a thinking support system called ThinkNavi, is scheduled to be released in the first quarter of 2026.
ThinkNavi can be used by individuals and organizations, and supports the formulation and planning of strategies and concepts through chat. However, it is not simply a chat system; behind the scenes, the ConceptMiner engine automatically organizes chat logs to build a concept network. This concept network serves as the foundation for defining strategies and concepts. Users do not need to acquire special analytical skills; simply communicate their thoughts through chat, and the LLM uses the appropriate functions in the concept network to obtain snapshot information and guide them toward defining strategies and concepts. Going forward, we plan to implement appropriate interfaces to keep up with advances in enterprise AI from generative AI providers such as OpenAI.