IP analysis platform PatSnap launches CoPilot, its new AI assistant

Founded in 2007 to help IP and R&D professionals do their jobs more efficiently, Singapore-based PatSnap‘s first product was a global patent search database. Now it’s building out its suite of AI products with the launch of its AI assistant CoPilot.

CoPilot lets users search patent and non-patent literature more quickly, which the company says will make IP and R&D workflows faster across PatSnap’s entire product suite. Its proprietary LLM also links back to sources, including journal and patents, and provides references.

The startup says it has spent millions of dollars and has a team of more than 50 engineers dedicated to its AI capabilities. PatSnap CEO and co-founder Jeffrey Tiong tells TechCrunch that PatSnap exists to remove friction in the innovation process for its customers, both within IP and R&D teams, and between them. IP analysts and attorneys need to run prior art (information publicly disclosed before a patent was filed) and freedom to operate (or the ability to use a product or service without facing legal issues) searches to decide where their companies should invest time and money, and PatSnap’s products are designed to make the process more smooth.

For example, PatSnap’s Analytics product includes over 180 million patents and over 130 million pieces of literature from 170 jurisdictions. IP teams can use PatSnap’s AI tools to analyze their markets and protect inventions at scale, says Tiong.

What CoPilot does is build further onto PatSnap’s AI products. It enables IP and R&D teams to find what they need more quickly within patents, non-patent literature and technical news. Some examples Tiong gives include getting automatic summaries of patent claims, learning about a technology and getting links to a company’s patents. It also answers specific problems. For example, if a R&D professional wants to know more about improving battery energy density, CoPilot can translate patents and find relevant literature.

CoPilot also helps R&D and IP teams in four other ways: keeping up-to-date with rapidly changing sectors, providing content analysis to help guide strategic patents and research, extracting key details from specific patents and literature, and AI security because its model isn’t trained on customer data. Tiong says that since CoPilot has its own proprietary LLM, customer data doesn’t need to leave PatSnap’s firewall, so it isn’t passed to external networks.

Tiong adds that PatSnap’s LLM was trained on data from patents, academic papers, technical reports and recent company news, including about mergers and acquisitions. It also incorporates data annotated by IP experts and PatSnap’s products.

Its LLM learns generic data, specialized data and aligned data in three stages: pre-train, post pre-train and self-training fine-tuning, so it has specialized accuracy in patent and non-patent data. Combined with retrieval-augmented generation, Tiong says that compared to GPT3.5, CoPilot’s model is better at performing in-depth analysis while being less likely to hallucinate answers.

PatSnap has raised $350 million from investors like SoftBank and Tencent. It has a total of more than 1,200 employees, with 12,000 customers. Its clients come from verticals including life sciences, automotive, consumer goods, technology, manufacturing, engineering and legal.