IPRally: Making the search for global patent records simpler with Google Cloud
About IPRally
IPRally is an AI startup based in Helsinki with more than 30 employees. The company’s web-based search application leverages the power of machine learning to enable customers to search for patents more easily. The company’s mission is to make technical knowledge universally accessible and usable.
Tell us your challenge. We're here to help.
Contact usIPRally leverages the power of Google Cloud to pinpoint patent results more precisely, while highlighting key information to help customers find what they’re looking for more quickly and easily.
Google Cloud results
- Updates around 4 million documents and adds an additional 200,000 searchable records each week
- Leverages the power of multiple processing clusters in five different zones across three global regions to ensure optimal performance
- Reduces cost of graph parsing runs by more than 50%
- Delivers relevant search results in under three seconds while highlighting key contextual information
Turns 100 million patent documents into knowledge graphs
Before any aspiring inventor tries to bring their big new idea to market they need to know that someone else hasn’t got there first. But finding out if your idea is actually original can be a time-consuming challenge. Innovators, inventors, and thought leaders have long been faced with the prospect of scouring existing patents and other intellectual property rights (IPR) records to ensure that their latest brainchild is genuinely unique.
Traditional digital approaches to patent searching use text-based queries to find results for existing technologies and methodologies. While they can deliver tens of thousands of results, patent professionals are still left to sift through the data to check for relevant matches. But a new web application, IPRally, has changed the way many of the largest operators in the IPR space do their patent searches.
IPRally leverages the power of Google Cloud to generate knowledge graphs and train graph neural networks to help pinpoint relevant results much more precisely. And its algorithms are able to highlight key information within these results to help people find what they’re looking for quickly and easily. All of which results in an intelligent search engine that operates like a patent professional and delivers results in under three seconds.
When it came to building IPRally, its creators evaluated other cloud providers, before deciding that Google Cloud was the right fit. "We needed advanced machine learning capabilities to create IPRally, and Google Cloud really stood out from the crowd in this area," says Juho Kallio, Co-Founder and CTO, IPRally. "We were also hoping to attract interest from large global clients so we needed a cloud platform that could meet this level of demand."
"We needed advanced machine learning capabilities to create IPRally, and Google Cloud really stood out from the crowd in this area. We were also hoping to attract interest from large global clients, so we needed a cloud platform that could meet this level of demand."
—Juho Kallio, Co-Founder and CTO, IPRallyBuilding IPRally from the ground up in Google Cloud
IPRally took its first steps toward success in 2018, creating a POC patent crafting tool. Although this was never released for public use, it was enough to lead to funding that allowed IPRally to begin building its search engine, a process that took 18 months. Among the key innovations that IPRally struck upon was a process called graph parsing, which uses machine learning to turn raw text into knowledge graphs. Deep learning is used in graph parsing, and also with the graph neural networks that work with the graphs. With the unique graph approach, IPRally has created a more accurate and user-friendly patent search engine. The user mainly sees the search document with AI highlights, while the knowledge graph is hidden and helps the tool with finding and highlighting. This means users don’t have to trawl through pages and pages of dense technical information.
IPRally originally built these services in App Engine, utilizing microservice infrastructure to separate the various processes involved, with a CI/CD pipeline in place from the very beginning. This approach afforded IPRally the scalability it needed to adapt to fluctuating demand for its services, as well as the flexibility to use different coding languages for its frontend and backend infrastructure (Clojure), as well as its graph parsing and graph neural networks (Python).
Scaling up to meet current and future demand
As IPRally improved its deep learning models and grew from its initial phase of deployment, it needed to update its technology stack accordingly. The company decided to migrate its application components to Google Kubernetes Engine containers, using Pub/Sub to push processes to preemptable Compute Engine VM instances on node pool clusters to ensure fault-tolerant workloads.
This was only possible once IPRally had a proper understanding of its current needs and likely future demand. To date, IPRally has converted 100 million global patent documents into searchable knowledge graphs. And the service is kept up to date with the addition of around 4 million updates to existing documents, as well as the addition of 200,000 searchable records, each week. All of this data, around 230TB in total, is stored in Cloud Storage, while a PostgreSQL database of around 3TB is used as the web application’s main permanent relational database.
IPRally says that the Google for Startups Cloud Program was a big help in achieving this increase in scale with minimal initial funding. "Being given free credits had a huge impact on our initial growth," says Juuso Piskonen, Co-Founder and Engineering Lead, IPRally. "The sort of scale we’re talking about is really expensive for your average startup with minimal funding. Once we got everything scaling we were able to get more funding, which allowed us to continue to grow further and properly plan for our future."
"Solutions like Cloud GPU are enabling us to stay at the bleeding edge, and that’s what we live for at IPRally. The faster we can run our graph parsing processes, the more time we can spend working to provide new data sets and deliver additional value to our customers."
—Juho Kinnunen, Lead DevOps Engineer, IPRallyOptimizing performance to increase data processing
When it comes to data processing, including its parsing runs, IPRally is looking for even more processing power. With the peak load, when the periodical knowledge graph parsing is happening they run on top of around 400 Google Cloud GPUs and 8,000 CPUs located in five different zones, and across three global regions, to ensure optimal performance. This has enabled IPRally to modernize the neural network technology used for graph parsing. The new neural networks bring IPRally closer to other data sources in addition to patents.
With these improvements in available data, IPRally can deliver increasingly accurate results to its customers, which the company says is one of its key selling points. The company has also been able to reduce the cost of graph parsing runs by more than 50%. This cost efficiency is crucial to the business, as it means that the money can be spent on research and development opportunities within the business.
"The faster we can run our graph parsing processes, the more time we can spend working on our algorithms and deep learning solutions, so we can deliver additional value to our customers," says Juho Kinnunen, Lead DevOps Engineer, IPRally.
Expanding IPRally to change information access
IPRally says that it’s proud of the rapid progress it’s made in such a short time. It credits much of its success to its decision to work within Google Cloud. "IPRally was a completely new concept when we created it," says Kallio. "And with Google Cloud, we’ve been able to build a product that our customers love, one that’s providing a lot of value to the world."
By working even more closely with Google Cloud, IPRally believes it can do even more great work in the future. "Our early architecture decisions are really paying dividends," Kinnunen says. "They’ve meant that we haven’t begun to lag behind the competition because of delays in the parsing pipelines. And, from a recent perspective, it's been really nice to work with Google Cloud engineers and our account manager to ensure that we’re getting the most out of our setup. It's been a really personal approach."
One area that IPRally is keen to explore is the world of scientific articles and R&D papers. The company wants to change the way that people access technical information. "We’ve come a long way," Kallio says, "but it still feels like we’re just at the beginning. It’s becoming more and more of a challenge to find relevant information. What if we could quickly provide customers with information on topics that have been researched previously? And AI could understand the information within a document even more precisely and quickly than a person? The possibilities are endless."
"IPRally was a completely new concept when we created it. With Google Cloud, we’ve been able to build a product that our customers love, one that’s providing a lot of value to the world."
—Juho Kallio, Co-Founder and CTO, IPRallyTell us your challenge. We're here to help.
Contact us
About IPRally
IPRally is an AI startup based in Helsinki with more than 30 employees. The company’s web-based search application leverages the power of machine learning to enable customers to search for patents more easily. The company’s mission is to make technical knowledge universally accessible and usable.