The extraction of tabular data from documents presents significant challenges, especially with diverse table layouts, structures, and content. Existing Intelligent Document Processing solutions address some aspects of table processing; however, many fail to deliver a comprehensive approach that combines high accuracy, scalability, and adaptability. This paper proposes an integrated Optical Character Recognition and Large Language Model framework designed to address tabular processing challenges and improve intelligent document processing. A custom data set, comprising tables from academic papers and business documents, is created to evaluate the framework’s performance. Each table is paired with two questions, ensuring a comprehensive evaluation of the framework’s ability to accurately extract and process information from a variety of table types. The results indicate that the proposed framework performs well in extracting data across different contexts and table formats, with room for further optimisation. The highest accuracy was achieved using structured JSON files without prompt engineering, highlighting the impact of preprocessing choices on performance. This work contributes significantly to the field of intelligent document processing and offers a foundation for future research in table extraction and understanding.