SciPat - Scientific Patents

Currently success rates for patents are 54% and 65% in the US and Canada respectively while filing costs are up to 15-20k per patent. With over 600000 patents being filed annually in the US (2019) and over 12000 in Canada, around 2.5 billion dollars is being spent on patent applications that end up failing. Furthermore, the process to file a patent is time intensive and requires legal experts and technical experts to collaborate to create high quality patents. Of the 600000 patents filed in the US, ~20% of are for inventions classified as deep tech in fields such as materials science, pharma, semi-conductors, etc. Patents in these fields are generally complex as strong expertise is required for prior art searches and for defining embodiments during patent generation. Current solutions are generalized to enable lawyers to perform prior art searches and to generate for fields of patents, however we believe there is an opportunity to build specialized tools specifically for deep tech fields where patent application costs are higher. Therefore we've started developing SciPat (scientific patents) a set of tools to help both lawyers and inventors in these spaces to perform specialized prior art searches and generate specialized drafts. To accomplish this we are developing a set of specialized models by fine-tuning cohere's generative model, embedding model and summarization model using field specific patents and literature while also providing the base functionality provided by other generative patent and prior art search tools.