Introduction

At IPRally, we solve problems in four main focus areas: patent search, review, monitoring, and classification. Each of these functionalities is associated with numerous use cases in the industry. They are frequently interconnected to such an extent that we refer to them collectively as "patent understanding," encapsulating the broader underlying challenge.

We are confident to say that we have the most powerful patent search engine in the world due to substantial investments in our graph AI search technology. But as everyone in the industry knows, the problem is hard to solve and no AI search system is flawless. Moreover, deep content reviews are needed in other areas, such as portfolio analyses – or, at the very least, they would prove advantageous.

The review process has traditionally been enhanced through a number of approaches we’re all familiar with: user friendly presentation of the data, highlighting, smart image views, manually or automatically enriched data, document annotations, etc. All of this is used in the industry and most of it is available in IPRally too. 

But until now we have seen very few efforts addressing the core of the problem: the actual understanding of the technical substance of patents, which is often very detailed and certainly diverse.

Now pieces of the patent understanding puzzle appear to be falling into place. Last year generative AI reached a true "human assistant level" and sometimes beyond in many areas. Patent understanding, in the true sense of the word, was born. 

Below, I outline what we at IPRally have done so far to help understand patents better and faster -  as well as what’s coming next. 

Ask AI

Last September, we launched our inaugural generative AI feature, Ask AI. Since its release, the feature has gained significant traction and has proven immensely helpful in extracting technical details and verifying the presence of specific features within individual patents.

The advantage lies in the AI's robust repository of common general knowledge, its ability to grasp context, and its capacity to address inquiries and commands without requiring users to meticulously frame their questions. Just ask! 

Another task at which Ask AI excels is the contextual summarization of patents. Requests such as "Summarize the material strength-related aspects of the patent", "Condense each claim into simple terms", or "List the main embodiments and their key advantages" pose no challenge for Ask AI.

Ask AI proves immensely helpful not only in conducting thorough reviews of search results but also, for instance, in facilitating communication with non-patent professionals within the organization. “Finally, patent language can be easily converted to human language”, remarked one of our users.

When it comes to users, they've demonstrated remarkable creativity in utilizing Ask AI. This was both anticipated and desired when we decided to develop the feature in a flexible manner and give control to the user.

Multi-Patent Ask AI 

Multi-patent Ask AI extends the capabilities of Ask AI to handle large sets of patent publications. Users can pose identical questions or issue commands for a few, tens, or even hundreds of patents simultaneously. The responses are delivered essentially in real time.

This enables the expansion of AI-assisted review to longer result lists and even entire patent portfolios. Users can effortlessly refine search results to uncover the most crucial hits.

Multi-patent Ask AI in action in reviewing search results. First, the feature is used to identify a few highly relevant favorite hits, and then to drill deeper into those hits.

‍

Use Case Examples

1. AI assisted analysis of search results‍

Pose the following question to the hits of your semiconductor search:  

“Does the patent mention passive temperature compensation of the semiconductor element? If so, which compensation mechanisms are disclosed?”.

Within seconds, you'll receive a clear and concise answer for all your search results, readily available for your immediate use. Whether you need to dismiss irrelevant results or identify the most promising ones for your refined search, they'll be right at your fingertips. 

How much time would it take to achieve the same outcome manually, by opening each document individually and reading its contents? Soon, we'll simply want to let it fade from memory and not look back. 

2. AI assisted portfolio insights

Take your own, your competitor’s or your acquisition target’s portfolio and ask: 

“Does the patent disclose or include any hints of using synthetic biofuels? Which ones?”

A technical field that has been developing for such a long time is very crowded, and the patents contain a lot of secondary and even hypothetical embodiments. In such instances, it is nearly impossible for anyone to gain a comprehensive understanding of the aspects disclosed in the patents, let alone determine whether they fall under the claims.

Multi-patent Ask AI can quickly sift through a vast number of documents, each spanning tens of pages, to deliver focused answers in a digestible format. A task that was once out of practical reach or demanded significant and tedious human effort is now entirely achievable.

This opens up new opportunities for getting strategic and competitive insights, and more informed IP decision-making. 

Can I Rely on the Answers?

We are paying a lot of attention to the factuality of the answers provided by the AI. This is achieved through three main approaches: 

• Using reliable AI models
• Making the models operate in a “non-creative mode”
• Instructing the models to stick to the contents of each individual patent, without surpassing the disclosure, while also considering common general knowledge in the industry – just like a human would do

This being said, we cannot guarantee that each of the answers would be fully correct. Occasional hallucinations of LLMs are a known fact and knowing the tools used, as well as double-checking the facts from the original document before making eventual decisions, is and will be crucial.

For optimal performance, we recommend using relatively short questions with not too many focuses.

However, user feedback overwhelmingly attests to the usefulness of the answers. The recurring questions are: How would a human perform the same task? What value do insights derived from AI offer compared to having no insights at all?

When and How Can I Use it?

The single-patent Ask AI feature is available for all customers and trial users, included in IPRally at no extra cost. As for Multi-Patent Ask AI, 1000 free credits per organization are provided to customers for evaluation purposes. Free trials include 100 free credits.

Register for a free trial or contact our sales at sales@iprally.com to get in touch.

‍Read the press release here.

‍

‍