In 2007, PatSnap was founded in Singapore with a mission to enhance the efficiency of IP and R&D professionals. Their initial offering was a comprehensive global patent search database. However, as the company continues to grow, they are expanding their AI tools with the introduction of CoPilot, their highly anticipated AI assistant.
“CoPilot allows for faster and more efficient searches of both patent and non-patent literature, streamlining IP and R&D workflows across all PatSnap products,” boasts PatSnap CEO and co-founder Jeffrey Tiong in an interview with TechCrunch.
CoPilot’s innovative LLM technology links back to sources such as journals and patents, providing valuable references for users.
PatSnap has invested a considerable amount of resources, totaling in the millions, and has a dedicated team of over 50 engineers solely committed to advancing their AI capabilities. According to Tiong, PatSnap’s ultimate goal is to eliminate barriers and challenges in the innovation process for their customers, simplifying communication and collaboration within and between IP and R&D teams.
For IP analysts and attorneys, conducting prior art and freedom to operate searches is a crucial aspect of determining where a company should focus their time and investments. PatSnap’s suite of products is designed to streamline this process and make it more efficient.
For instance, their Analytics product boasts an impressive database of over 180 million patents and 130 million pieces of literature from 170 different jurisdictions. Tiong explains that the AI tools offered by PatSnap enable IP teams to analyze their respective markets and protect their inventions on a larger scale.
“CoPilot builds upon PatSnap’s existing AI products, providing IP and R&D teams with answers at lightning speed from within patents, non-patent literature, and technical news,” says Tiong.
CoPilot offers a wide range of features to aid R&D and IP professionals in their work, including automatic summaries of patent claims, links to relevant patents for technology research, and even assists with specific problems. For instance, CoPilot can assist an R&D professional looking to improve battery energy density by translating patents and finding related literature.
CoPilot’s capabilities extend beyond just aiding in research and analysis. Tiong highlights four additional ways CoPilot can benefit IP and R&D teams. These include keeping up-to-date with rapidly evolving sectors, providing content analysis to guide strategic patents and research, extracting key details from specific patents and literature, and ensuring AI security by training their model on their own data instead of customer data. As Tiong explains, CoPilot’s proprietary LLM eliminates the need for customer data to leave PatSnap’s firewall, reducing the risk of data being passed to external networks.
Tiong goes on to describe how PatSnap’s LLM was trained on data gathered from various sources such as patents, academic papers, technical reports, and recent company news, including details on mergers and acquisitions. Additionally, the LLM incorporates data annotated by both IP experts and PatSnap’s own products.
The LLM’s training is divided into three stages: pre-train, post pre-train, and self-training fine-tuning. This approach enables the LLM to learn generic data, specialized data, and aligned data, giving the model a high level of accuracy in both patent and non-patent data. Combined with retrieval-augmented generation, Tiong asserts that CoPilot’s model outperforms even GPT3.5, offering in-depth analysis without the risk of providing incorrect or “hallucinated” answers.
In terms of financial backing, PatSnap has raised an impressive $350 million from investors such as SoftBank and Tencent. With a team of over 1,200 employees and a customer base of 12,000, PatSnap serves clients across various industries, including life sciences, automotive, consumer goods, technology, manufacturing, engineering, and legal.
