Introducing Smart filters: redefining patent review

IPRally’s mission is to automate tedious, substance-level patent work using AI – efficiently, reliably, and in a verifiable manner. Efficiency and trust are at the core of everything we do.

I’ve previously outlined our long-term vision in this blog, and today, I’m excited to introduce Smart Filters, our latest feature that sets a new standard for efficient patent dataset analysis. This is a major step toward full automation, significantly accelerating patent reviews. At best, it reduces review time to a fraction of what it once was; at minimum, it delivers a substantial efficiency boost. All of the pains that come with classical filtering are replaced by a simple, conversational and proactive assistant that works just like an extension of your patenting team.

What are Smart filters?

In a nutshell, Smart filters:

• Help to narrow down patent datasets intuitively by asking questions about them.
• Make our Multi-patent Ask AI feature directly actionable, by 
  
  • detecting if the user asks a Yes/No question, 
  • answering the question based on each document, including both reasoning and a final answer: “Yes”, “No” or “Maybe”, and 
  • allowing users to filter their results set using those answers.
• Utilize a large language model (LLM) in "patent professional mode", to generate the answers individually from each document. 

This feature is particularly helpful in the following use cases:

• Feature-level analysis of novelty, invalidity and state-of-the-art search results, for example
• Understanding of and in-depth analysis of patent portfolios
• Office action response work and defensive invalidity and opposition work

There is a video overview at the end of this blog, after a few example questions and use cases to spark some ideas on how you can use it to solve your day-to-day challenges.

Difference from classical Boolean filtering

Smart filters have significant advantages over traditional keyword and patent class-based filtering, as they leverage the built-in language and engineering understanding of patent-trained LLMs.

Imagine, for example, a common challenge in the sports watch domain of finding patents that discuss using outputs of multiple sensors (e.g. GPS, accelerometer, atmospheric pressure) combined to determine altitude accurately in varying conditions. 

Since a lot of sports watch patents list multiple sensors, with varying terminology, as well as multiple parameters derived from them, it is impossible to capture such technical concepts with boolean queries. Any search or filtering efforts will, symptomatically, contain a lot of noise, and are likely to miss relevant documents.

Smart filters solve this problem by using Ask AI to simply ask: “Does this patent disclose using outputs of multiple sensors combined to determine altitude?”. The LLM then analyses each patent carefully, finds the relevant disclosure, provides reasoning why it thinks the disclosure is or isn’t there, and summarizes the answer in a simple and actionable “Yes”, “No”, or “Maybe”.

Another illustrative example is the use of a substance: “Does the patent disclose the use of ammonia as fuel?” in an area where ammonia has multiple uses is trivial to answer with Smart filters, but very difficult to answer with classical query methods. 

In short, Smart filters remove the need to define lists of synonyms or create complex boolean queries with wildcards, proximity operators etc, as well as reduce the reliance of IPC/CPC classifications. It can perform tasks that are practically impossible with Boolean. 

All of the pains that come with classical filtering are replaced by a simple, conversational and proactive assistant that does the job in no time.

Use case examples

Consider having tens or hundreds of patent search results that you need to review. With Smart Filters, you can directly address the most relevant aspects of your search problem either with single-aspect questions or multi-aspect questions:

Exemplary single-aspect Smart filter questions

• "Does the document disclose a piezoelectric transducer suitable for pressure sensing?"
• "Is there an active temperature drift compensation mechanism?"
• "Does the document disclose kaolin as filler in the main layer of the sheet?"
• "Does it discuss the relationship between the temperature of the lubricant and energy efficiency of the apparatus?"
• "Is it a multi-stage drying process?"
• "Does it address the problem of reducing vibrations at low rotation speeds?"

Single-aspect questions focus on one feature at a time, but as you can see, the aspect can be a complex technical concept, with structural, functional or purpose-related sub-aspects. A typical answer looks like this:

Exemplary multi-aspect Smart filter questions 

Does the patent disclose all of the following features:

A) A material suitable for use in a three-dimensional printing system;

B) a carrier ink containing pigment particles;

C) the pigment particles comprising mica; and

D) a colorant that is different from the pigment particles.

Analyze feature by feature, indicate potential gaps, and cite the relevant passages for each feature. 

Multi-aspect questions can contain multiple or even all features of an invention.

Note that each question above can in principle be answered in a binary way: “Yes” or “No”. For granularity, we also give a “Maybe” answer if the AI is not sure about the answer. 

Answers to single-aspect questions are generally more informative and accurate, and allow step-wise filtering or building a comprehensive feature chart for a claim, for example. However, in a knockout-type search, a multi-aspect question can take you directly to one or more highly relevant hits, and in a best-case scenario, solve your case with a single question.

Questions that are better to ask in an open format

In certain cases, it is still better to just extract the information from the documents and draw conclusions yourself. For example, LLMs still struggle with numbers and unit conversions, and are not fully reliable e.g. in analyzing range overlaps. Yet, they are very good at extracting the ranges from documents.

Consequently, it is better to ask Ask AI "Which temperatures or temperature ranges are disclosed?" and manually review the answers rather than ask "Does this patent disclose a temperature range of 80-120C?" and then utilise Smart filters. 

It is worth noting that descriptive questions and commands, like summarization requests, can be made using the same interface. The rule of thumb is: just ask what you need to know, we’ll take care of the rest.

Video demo

Additional notes and tips

• Double-check the answers: Smart filters are based on LLMs, which are powerful, but not perfect. While we are taking measures to prevent it, they can still miss details, misunderstand the question or misinterpret the documents. Also, numerical paragraph references can sometimes be slightly off. 
• Analysis of short patents is more reliable than long patents. The possibility of missing information or referencing wrong paragraphs increases e.g. with long pharmaceutical patents.
• Smart filters cannot see image data: Answers are based on the text content of patents only. Tables and chemical formulas are considered only if they are in text format and not as images in our document view.
• Smart filter answers, just like the normal Ask AI answers, are exportable in XLSX, DOCX or PDF formats, including both the short answer and reasoning.
• Smart filters process the documents individually in parallel. If you need aggregated analysis, such as a summary of answers, you can use your preferred LLM to create the second-stage analysis conveniently using the export functionality.

Next steps

As some of the next steps, we plan to enhance Smart filters by leveraging AI to automatically generate not only the answers, but also help you formulate the questions. We are also looking into automating the entire process, from the initial search to result analysis and a summary report. All in line with our mission to provide full automation of repetitive patent-related substance-level work. Meanwhile, we hope that Smart filters will make your work more enjoyable and efficient, and even unlock new use cases to make your R&D, intellectual property and business thrive.

We are keen to hear your feedback!

Join our upcoming Smart filters webinar

Sign up for our our introductory webinar to see how Smart filters works its magic and gain insights into best practices for your review process. 25 February 2025, 5 PMCET (11 AM ET). Welcome with your registration!

Try IPRally for free

Sign up for a free trial of IPRally and experience cutting-edge AI patent search, including features like Ask AI and Smart filters – completely free, no strings attached!