Rowan Patents continues to advance generative AI functionality in the Rowan Patents integrated drafting environment by combining our deep expertise in patent drafting software with the rapid advancements in Generative AI technology.

We remain focused on enhancing the efficiency of patent practitioners through the responsible use of AI within the patent drafting process. The new Rowan GenAI Assistants present practitioners with options based upon the drafting task to generate many parts of the application while at the same time giving the practitioner control over where to use GenAI. We apply drafting-specific prompt engineering to improve the utility of language model outputs to produce text for select parts of a patent application. Importantly, the GenAI Assistants leverage the integrated nature of Rowan Patents, maintaining linked objects such as terms and part references to ease revision of any AI-generated content and ensure consistency, which is critical to quality drafting.

The Rowan GenAI Assistants are powered by a local language model (Microsoft phi-3-mini) that is ideally suited to address the task-based use of AI for the unique needs of each section of a patent application. Using a local language model maintains security by keeping any patent matter the user inputs into the GenAI Assistant local to the user’s system. Rowan has created pre-built generative AI workflows for each assistant that are also stored and run locally. No data is transmitted out of the system for any part of these generations.

The initial release uses claim language to generate text for areas of the specification, including the Title, Technical Field/Field of Invention, Background, Brief Summary, and Abstract. Importantly, the system uses prompt engineering to guide the model in delivering content based upon statutory requirements and patent drafting best practices that are appropriate to the section of the application being drafted. For many tasks, a custom validation pipeline identifies and addresses potential issues before the generated text is inserted into the document. For example, when generating an abstract, the pipeline can detect when the word count is outside the allowable range and iteratively re-prompt the model to revise the text to be more concise until it falls within the range.

By taking this approach to the new Rowan GenAI Assistants, we remain focused on enhancing the efficiency of patent practitioners by balancing the responsible use of AI with the human oversight and expert judgment of a practitioner critical to efficient and high-quality patent drafting.