Using AI and Neural Networks for Patent Searches: A Blueprint for Technological Advancement
Every year, tens of thousands of patent attorneys use millions of hours for defining inventions in natural language documents. At the same time, thousands of patent examiners spend millions of hours looking for novelty factors within those documents. Enter IPRally and its AI technology, which goes deeper than keyword searches to truly learn the logic of patenting.
We spoke with Sakari Arvela, CEO and Co-founder of IPRally, about his high hopes for graph neural networks; what this means for patents, tech and beyond; and why it’s so important to find the right investor partners who believe in the long-term outcomes of R&D.
First off, a very basic question: What global problem or inefficiency are you solving at IPRally?
The patent field is very traditional, and things have been done the same way for 100 years. Patent searching was, in the old times, done by going to a library and starting to crawl papers and find the relevant information. The next improvement was when computers came 30–40 years ago, and then it was possible to start making automatic keyword queries to patent databases — like “car” or “engine”. But since that, very little has happened, and kind of that makes the problem pretty interesting. There is such a strong, established way of doing things. We’ve been able to offer something new to this field: a way that resonates with how patent experts structure information in their heads and in spreadsheets. We have automated much of that.
Why is it valuable for patent search technology to go beyond just keyword search and into human-like processing patterns?
Patents are all about technology, and technology is all about interconnection. This bolt engages with this nut, and then it builds the machine. It’s all about functions and interactions. Those kinds of things are very hard to tell to a machine so that the machine can understand them. Now we have built a knowledge graph-based approach, where this kind of relational information can be told to the machine.
There is a branch in the machine learning field called graph AI, or graph neural networks. There’s a pretty good match between how we model technical information and what these graph neural networks can do. Graph neural networks offer real breakthroughs in other domains too, like DNA research, weather research or modeling very complex things. We have high hopes and expectations on how the graph neural networks can can solve these problems.
What is your approach to fundraising when you have such an R&D-heavy and technical product?
We must prioritize both our current product and long-term R&D, research and tech investments. There’s a balance of offering a product that serves our customers right now — with improvements that they want and need — while still keeping long-term focus clear year after year.
We are looking for an investor that believes in our long-term vision, so that we can really take it to the next level and make those breakthroughs that we have seen in many other fields like, AI image generation. When we reach that, then it can have a huge impact on the whole field — not only how patents are found and searched for, but also how they are understood and how quickly their information reaches the eye of an engineer or patent professional.
In short, we look to find investors who believe in us right now and in the future. On top of that, we are always looking to find new customers whose needs match with our current offering and who want to join the journey with us.
What role has Join played in your journey in terms of accelerating growth as an AI company?
Join’s networking in the early stage was very helpful for us, as they made many intros to the European markets. Many of those introductions led to those companies becoming our current customers. Also, like any good investor and board member, Join pushes us, gives us ideas, challenges our existing methods and makes us think. That’s always important.
For more information on IPRally, visit the company’s website.
