Established back in 2007 with the mission to enhance the efficiency of IP and R&D professionals, Singapore-based PatSnap has evolved from its origins as a global patent search database to introduce its latest AI assistant, CoPilot. The platform aims to accelerate patent and non-patent literature searches, facilitating faster IP and R&D workflows across Patsnap’s comprehensive product suite.
PatSnap’s CEO and co-founder, Jeffrey Tiong, shared that the startup is committed to eliminating friction in the innovation process for its customers, streamlining collaboration within IP and R&D teams. The company has invested heavily, dedicating a team of over 50 engineers and spending millions of dollars to develop its AI capabilities.
CoPilot extends PatSnap’s AI products, providing IP and R&D teams with a tool to efficiently navigate through patents, non-patent literature, and technical news. Tiong highlighted several functionalities, including automatic summaries of patent claims, technology insights, and links to company patents. The AI assistant also addresses specific queries, such as translating patents and finding relevant literature on topics like improving battery energy density.
PatSnap’s Analytics product, boasting over 180 million patents and 130 million pieces of literature from 170 jurisdictions, enables IP teams to analyze markets and protect inventions at scale using AI tools. CoPilot complements these capabilities, offering a faster and more targeted search experience.
CoPilot contributes to IP and R&D teams in various ways, including staying updated on rapidly changing sectors, providing content analysis for strategic patents and research, extracting key details from specific patents and literature, and ensuring AI security by keeping customer data within PatSnap’s firewall.
Tiong emphasized the proprietary nature of CoPilot’s Language Model (LLM), which was trained on diverse data sources, including patents, academic papers, technical reports, and recent company news, incorporating annotations by IP experts and PatSnap’s own products. The LLM undergoes three stages of learning—pre-train, post pre-train, and self-training fine-tuning—ensuring specialized accuracy in patent and non-patent data. Tiong claimed that, compared to GPT-3.5, CoPilot’s model excels in in-depth analysis while being less prone to hallucinate answers.
PatSnap, having raised $350 million from investors like SoftBank and Tencent, boasts a workforce of over 1,200 employees and serves more than 12,000 customers across diverse verticals, including life sciences, automotive, consumer goods, technology, manufacturing, engineering, and legal. The introduction of CoPilot further strengthens PatSnap’s commitment to providing cutting-edge solutions for intellectual property and research and development professionals.