Cross-Domain R&D Concept Exploration System

Consider this if you are facing challenges in identifying R&D themes.

Searching through academic papers, patents, and technical news to identify new R&D themes is a standard approach. However, consider this: your search is inherently limited to what you already know by name or keyword. While tools known as “Technology Intelligence” (TI) have emerged recently, the fundamental limitation remains the same.

Isn’t the real reason you can’t find an R&D theme that your area of ​​expertise has become saturated? If new possibilities lie in industries or technological fields different from your own, you wouldn’t yet know the names of those fields or the keywords to search for, would you?

How do you search for something you don’t know exists? That is the essence of this problem.

ConceptMiner R&D—Cross-Domain R&D Concept Exploration is an AI system designed to solve this problem. By extracting the winning patterns, structures, relationships, and archetypes inherent in a company’s accumulated technologies and exploring them using Large Language Models (LLMs), the system provides clues as to where to focus exploration efforts. The process begins with a comprehensive inventory of the company’s technologies. While traditional approaches might have involved manually filling out survey forms, the modern method is different; it starts with the construction of a knowledge base. However, this is not merely a system that answers questions like a standard RAG (Retrieval-Augmented Generation) setup.

In short, ConceptMiner R&D automatically generates a knowledge base—structured around conceptual models—from internal technical documents and establishes an environment where an LLM can reference them. By extracting, abstracting, and generalizing key technical elements, it formulates “inquiries” that allow the LLM to explore the knowledge space it has already acquired from a cross-cutting perspective. While an LLM yields only mediocre answers to mediocre questions, it responds with its full capability when presented with questions framed from a sharp, insightful angle. LLMs possess detailed knowledge across all technical fields. You can then create a strategy map based on the answers obtained. If you identify a promising theme from a strategic standpoint, you can proceed to explore it using conventional methods and tools.

Create New Questions from Known Technologies

The “questions” generated by ConceptMiner R&D involve asking the following about technologies the company already possesses:

Where else could this principle be useful?

Which other industries require the same function?

What other fields face a structurally similar problem?

For example:

maintaining fine particles in a stable and uniform dispersion over long periods.

Instead of searching only by the current product name or application, we can ask:

In what other situations, outside the current application, would the principle of maintaining fine particles in a stable and uniform dispersion be valuable?

An LLM may then identify industries, applications, research fields, or technical terms that were not previously part of your search vocabulary.

Those newly discovered terms can then be used in conventional searches.

The process becomes:

Known technology
→ Abstracted question
→ Unknown candidate field
→ New search terms
→ Conventional investigation

This is how we build a bridge from known knowledge to unfamiliar domains.

The Starting Point Is Your Existing Technical Knowledge

Companies already possess large amounts of technical knowledge in forms such as:

  • R&D reports
  • technical papers
  • patents
  • design documents
  • experiment records
  • product documentation
  • past research themes
  • technical proposals
  • manufacturing and quality documents

These documents contain more than product names and technology labels.

They also contain:

  • materials
  • functions
  • principles
  • physical or chemical effects
  • processes
  • manufacturing methods
  • control technologies
  • measurement technologies
  • technical problems that have been solved
  • operating conditions
  • constraints
  • accumulated know-how

Any of these elements may become a starting point for cross-domain exploration.

A Modern Technology Inventory Can Begin with a Knowledge Base

When Fujifilm faced the rapid decline of the photographic film market, the company systematically reviewed the technologies it had accumulated through its photographic business.

Rather than viewing its capabilities only in terms of the final product—photographic film—it examined underlying technical assets such as collagen-related knowledge, antioxidant technologies, nanotechnology, and optical analysis and control.

These capabilities later contributed to new businesses including cosmetics and healthcare.

The important point is not that someone simply had a sudden idea to move from film to cosmetics.

The company first looked closely at what technologies it actually possessed and then considered where else those capabilities could create value.

Today, this kind of technology inventory can be supported by AI.

Instead of relying entirely on interviews, workshops, and manual document review, technical documents can be organized into a knowledge base that LLMs can reference across the organization.

But the purpose is not merely to organize information.

The knowledge base becomes the starting point for cross-domain exploration.

Cross-Domain R&D Concept Exploration Process

We currently envision the following six-stage process.

STEP 1 — Build a Knowledge Base from Technical Documents

Relevant R&D reports, patents, design documents, product materials, and other technical information are organized so that an LLM can reference them across documents.

At the same time, ConceptMiner can use GNG (Growing Neural Gas) to organize the information according to conceptual similarity.

This produces a Technology Map showing:

What technical knowledge does the company currently possess, and how is it related?

The knowledge base is therefore not used only to answer internal questions.

It becomes infrastructure for exploration.

STEP 2 — Extract Technical Elements with an LLM

The LLM refers to the knowledge base and extracts technical elements that may serve as starting points for exploration.

Examples include:

  • materials
  • functions
  • principles
  • effects
  • processes
  • manufacturing methods
  • control technologies
  • measurement technologies
  • technical problems
  • constraints

The objective is not simple keyword extraction.

We try to separate the underlying technical capability from its current product or application.

The key question is:

What is this technology fundamentally doing?

A single technology or document may produce several different exploration elements.

STEP 3 — Combine Technical Elements with Question Patterns

Rather than asking an LLM to freely generate ideas, we prepare multiple question patterns representing different ways of exploring across domains.

These question patterns are automatically combined with extracted technical elements to generate prompts.

For example:

Explore by principle

Where else could the principle of XX be useful besides YY?

Explore by function

Which industries or applications require the function XX outside the current field?

Explore by problem structure

Which other fields face problems that are structurally similar to XX?

Explore by technical capability

In what other markets could the capability to perform XX create competitive advantage?

Explore by constraint

Which other fields need to solve similar problems under the constraint XX?

The same technical element can therefore be explored from several different directions.

Cross-Domain R&D Concept Exploration does not assume that there is one universal ideation method.

The objective is to systematically test multiple perspectives.

STEP 4 — Explore Cross-Domain Candidates with LLMs

The generated prompts are used to explore possible connections with other industries and technical fields.

At this stage, the objective is not to produce a finished R&D theme.

The objective is to discover:

  • unfamiliar industries
  • new applications
  • new technology areas
  • new technical problems
  • new research themes
  • new terminology and keywords

In this process, the LLM is used not primarily as an answer generator, but as a tool for discovering:

What should we investigate next?

The resulting candidates remain hypotheses at this stage.

STEP 5 — Investigate and Validate with Conventional Technology Intelligence

Once a new field or theme has been discovered, conventional Technology Intelligence becomes essential.

We investigate the candidate using:

  • academic papers
  • patents
  • technology news
  • companies
  • products
  • competitors
  • market information

An LLM saying that a technology “may be applicable” is not sufficient evidence for an R&D decision.

We need to determine:

  • whether relevant technologies actually exist
  • which companies and research organizations are active
  • whether the company may have a genuine technical advantage
  • whether the application is technically plausible
  • whether meaningful market opportunities exist

Cross-Domain R&D Concept Exploration therefore does not replace conventional Technology Intelligence.

The roles are different:

Concept Exploration discovers what should be investigated.

Technology Intelligence investigates it in depth.

STEP 6 — Structure and Compare Candidates with a GNG Strategic Map

As exploration continues, tens, hundreds, or even thousands of candidate items may accumulate.

At that point, a new problem appears:

There are too many candidates to understand the overall exploration landscape.

ConceptMiner can then be used again to organize the investigated candidates according to conceptual similarity.

The first GNG-based Technology Map answers:

What technologies do we currently have?

The final GNG-based Strategic Map addresses:

In which directions could those technologies potentially expand?

The map can help visualize:

  • relationships among candidate themes
  • clusters of similar opportunities
  • adjacent areas
  • more distant cross-domain areas
  • unusual or isolated candidates
  • relationships between existing technologies and new opportunities

Other criteria can then be added, such as:

  • fit with internal capabilities
  • novelty
  • technical feasibility
  • strength of external evidence
  • market potential
  • competitive intensity

ConceptMiner does not decide which theme should be selected.

Its role is to provide a structured landscape that supports comparison, discussion, and further exploration.

From a Technology Map to a Strategic Map

ConceptMiner is used at two different stages.

Technology Map

The Technology Map represents the company’s current technical knowledge.

It helps answer:

What do we already know and possess?

This is the Known Space.

Strategic Map

The Strategic Map incorporates the fields, technologies, applications, and themes discovered through cross-domain exploration.

It helps answer:

Where might our technologies lead us next?

This is the Explored Space.

The objective is to expand from the Known Space toward areas that were not initially visible.

Exploration Is Iterative

Cross-Domain R&D Concept Exploration is not a one-shot brainstorming exercise.

The process is iterative:

Technical knowledge
→ Technical elements
→ Questions
→ Cross-domain candidates
→ Investigation
→ New knowledge
→ New questions

As exploration proceeds, unfamiliar technologies and terminology become familiar.

Those new terms become new search keywords.

Research may reveal new principles, problems, or applications, which can then become the starting point for another round of exploration.

What was unknown becomes known, and the newly acquired knowledge becomes the bridge to the next unknown.

There Is No Single “Winning Pattern”

Different companies will have different starting points for cross-domain exploration.

For one company, a distinctive core technology may be the strongest entry point.

For another, it may be a difficult technical problem the company has already solved.

Possible starting points include:

  • materials
  • processing technologies
  • measurement technologies
  • control methods
  • manufacturing constraints
  • customer problems
  • unique know-how
  • past successful combinations of capabilities

Analyzing a company’s historical “winning pattern” and applying it to another domain can be useful.

But it is only one possible route.

Cross-Domain R&D Concept Exploration focuses less on selecting one universal method and more on asking:

What kinds of questions can connect what we already know to domains we do not yet know?

From a Knowledge Base for Answers to a Knowledge Base for Exploration

Enterprise knowledge bases are usually designed to answer questions such as:

Have we investigated this before?

What is the specification of this product?

How did we solve this problem in the past?

These are valuable uses.

But technical knowledge can also be used in another way:

as the starting point for exploring knowledge outside the company.

In other words, the role of a knowledge base can expand from:

finding answers from the past

to:

creating questions for the future.

Early Access Program

While the core technology behind ConceptMier R&D is well-established, applying it to actual corporate environments involves an experimental aspect. We believe that a truly refined exploration method can only be achieved by fine-tuning the system based on user feedback. Therefore, we are recruiting early adopters under special terms.

We would like to examine questions such as:

  • Can useful technical exploration elements be extracted from internal technical documents?
  • Can prepared question patterns expand the exploration space effectively?
  • Can the process identify cross-domain candidates that would be difficult to find through simple LLM prompting?
  • Can those candidates be validated using external technical and market information?
  • Can GNG-based strategic maps help structure and compare a large set of candidates?
  • Can the process ultimately produce R&D themes worth further investigation?

Basic Project Flow

1. Build a knowledge base from technical documents
Including a GNG-based Technology Map.

↓

2. Extract technical elements with an LLM
Identify potential starting points for cross-domain exploration.

↓

3. Automatically generate prompts from technical elements × question patterns
Explore the same technology from multiple perspectives.

↓

4. Explore cross-domain candidates with LLMs
Discover new applications, technologies, industries, research themes, and keywords.

↓

5. Investigate and validate with conventional Technology Intelligence
Use papers, patents, companies, technology news, and market information.

↓

6. Structure and compare the results using a GNG Strategic Map
Evaluate the overall exploration landscape and identify promising directions.

The cycle can then be repeated as necessary.

Expected Outputs

Depending on the project, possible outputs may include:

  • a technical knowledge base for the selected domain
  • a GNG Technology Map of existing technical assets
  • extracted technical exploration elements
  • question patterns used for exploration
  • discovered cross-domain fields and applications
  • newly identified search terms
  • validated papers, patents, companies, technologies, and market information
  • a GNG Strategic Map of candidate opportunities
  • prioritized areas for further investigation

The objective is not to stop at:

“The AI generated an interesting idea.”

We want to be able to trace:

Which internal technical asset produced the question?

Why did that question lead to the new field?

What evidence exists in that field?

How does the opportunity compare with other candidates?

Who This Project Is For

This project may be relevant if your organization wants to:

  • discover new R&D themes
  • identify new applications for existing technologies
  • explore opportunities beyond the current industry
  • make better use of accumulated technical documents and research results
  • use internal knowledge bases for innovation, not only retrieval
  • move beyond searches that repeatedly return information from familiar domains
  • turn LLM-generated ideas into a structured exploration process
  • review and reassess existing technical assets
  • improve the stage before Technology Intelligence: deciding what to investigate

From What You Know to What You Have Not Yet Discovered

You do not need to know the future in advance to explore new R&D themes.

You need a way to generate the next question from the knowledge you already possess.

Build a technical knowledge base.
Extract technical elements.
Generate questions from multiple perspectives.
Use LLMs to explore unfamiliar fields.
Investigate the newly discovered areas using conventional methods.
Then look at the entire exploration landscape through a strategic map.

The goal is to expand beyond the limits of known keywords and familiar domains.

Cross-Domain R&D Concept Exploration Early Access Program

If you are interested in testing this approach with real technical themes or internal technical documents, please contact Mindware Research Institute.

Contact Us / Request Project Information