This article aims first and foremost to provide a comprehensive overview of existing methodologies for patent data extraction, with particular attention given to the approach developed by Guillaume Guarino, known as Summatriz. By critically analysing the strengths and limitations of this method, we seek to highlight the challenges that persist in effectively extracting and structuring information from patent documents, challenges which remain central to the advancement of innovation intelligence and intellectual property analytics. Building upon this diagnostic, we present our own pipeline and methodological framework, designed as an enhancement to the current state-of-the-art. We detail the specific modifications and additions we introduce and articulate how these changes will address the shortcomings previously identified. Our proposal is founded on the premise that leveraging large language models (LLMs) will substantially enhance the accuracy, adaptability, and scalability of patent data extraction systems.