XLSCOUT Unveils ParaEmbed 2.0: a Powerful Embedding Model Tailored for Patents and IP with Expert Support from Hugging Face

This is a guest blog post by the XLSCOUT team.

XLSCOUT, a Toronto-based leader in the use of AI in intellectual property (IP), has developed a powerful proprietary embedding model called ParaEmbed 2.0 stemming from an ambitious collaboration with Hugging Face’s Expert Support Program. The collaboration focuses on applying state-of-the-art AI technologies and open-source models to enhance the understanding and analysis of complex patent documents including patent-specific terminology, context, and relationships. This allows XLSCOUT’s products to offer the best performance for drafting patent applications, patent invalidation searches, and ensuring ideas are novel compared to previously available patents and literature.

By fine-tuning on high-quality, multi-domain patent data curated by human experts, ParaEmbed 2.0 boasts a remarkable 23% increase in accuracy compared to its predecessor, ParaEmbed 1.0, which was released in October 2023. With this advancement, ParaEmbed 2.0 is now able to accurately capture context and map patents against prior art, ideas, products, or standards with even greater precision.

The journey towards enhanced patent analysis

Initially, XLSCOUT explored proprietary AI models for patent analysis, but found that these closed-source models, such as GPT-4 and text-embedding-ada-002, struggled to capture the nuanced context required for technical and specialized patent claims.

By integrating open-source models like BGE-base-v1.5, Llama 2 70B, Falcon 40B, and Mixtral 8x7B, and fine-tuning on proprietary patent data with guidance from Hugging Face, XLSCOUT achieved more tailored and performant solutions. This shift allowed for a more accurate understanding of intricate technical concepts and terminologies, revolutionizing the analysis and understanding of technical documents and patents.