6. AI assistants

Assistants are predefined, domain-specific queries to the underlying Large Language Model (LLM). They are used to execute specific questions, analyses or evaluations directly, reproducibly and within the respective search context.

INTERGATOR Patent Search is delivered with a small but growing number of standard assistants. These assistants support typical work steps in patent search – such as analyzing individual patent documents, comparing them with a reference document, evaluating results or creating structured reports.

Assistants are to be understood as tools that support the search process. They help users understand technical content more quickly, extract features, identify relationships and prepare results in a traceable way. However, they do not replace technical or legal review.

Assistants can be used in several places in INTERGATOR Patent Search. Depending on where they are used, they refer to different data and serve different purposes.

In the result list, assistants support working with multiple search results or with the current result list. For example, they can help prioritize relevant results, evaluate results with regard to a specific question or prepare further search and analysis processes.

Typical use cases in the result list include:

• starting specialized search assistants from a result,
• searching for similar documents,
• creating semantic document filters,
• preparing a relevance assessment,
• supporting the refinement and prioritization of the result list,
• as well as selecting relevant documents for further analyses or reports.

In the document preview, the assistants refer to the currently opened patent document. They support the content-related analysis of an individual document, for example through summaries, feature extractions, translations or technical analyses.

In the user interface, assistants are selected via the “AI tasks” dropdown menu. This menu contains the assistants available for the document preview.

Typical use cases in the document preview include:

• free questions or instructions relating to the opened document,
• technical brief descriptions,
• summary and analysis of claims,
• extraction of claim features,
• expansion of features using the description,
• derivation of synonyms,
• extraction of technical parameters with values and units,
• analysis of the innovation,
• creation of a component graph,
• translation of title, abstract or brief description,
• as well as comparative analyses with reference to a reference document.

Assistants can also be used for creating and preparing reports. Reports are used to document analysis results, technical assessments and search results in a traceable way.

A distinction must be made between publication-related reports and search-related reports:

• Publication report: refers to a single patent document and prepares central information, technical content, features, text references and analysis results for this publication in a structured way.
• Search report: refers to a search, a result list or a selection of relevant documents and documents, for example, the search strategy, filters, relevant results, assessments and technical relationships.

Not all assistants are available in every situation. Some assistants require a specific search context in order to be executed meaningfully.

In particular, comparative assistants require a reference document or a suitable document filter in the current query. This applies, for example, to assistants that compare a currently opened patent document with a reference document or evaluate the technical proximity between a result and a source document.

A reference document is typically created in an assistant-supported workflow, for example by:

• entering a publication number in a validity search,
• starting an FTO search with a source document,
• starting a monitoring search with a reference document,
• or using a semantic document filter in the current query.

If the required context is missing, the corresponding assistants may be grayed out, disabled or not executable in a meaningful way. In this case, a suitable search assistant should first be started, a reference document should be included or a semantic document filter should be set.

Assistants always work within the respective context of the current search, result list or document preview. For reliable results, it is therefore essential that the underlying query, the reference document and the selected documents match the technical question.

• Assistants are always based on a query to the LLM.
• Assistants are to be understood as tools that support the search process, not as a replacement for independent technical or legal review.
• The UI label “AI tasks” refers to the dropdown menu in the document preview through which document-related assistants are selected.
• Some assistants are context-dependent and require a reference document, a suitable source query or an active document filter.
• Detailed application examples can be found on the respective subpages.