The finance industry operates with a huge volume of information, that are stored as documents in both a structured and unstructured manner. The industry will fundamentally evolve by the development of a computational system that can extract knowledge from this vast set of documents, link them to the existing databases and store them in a form, which can be deployed for a large variety of professional tasks.

The extraction of knowledge from these documents revolves around two key concepts - understanding their structure and exploring the semantics of the language in these documents. This talk will highlight:

• The importance of understanding the visual structure of a financial document, and how it enables downstream knowledge extraction tasks.
• Various knowledge extraction techniques, using a combination of transformer-based and classical NLP techniques, which are currently being deployed to address significant real-world problems.
• The importance of correct reading order from irregular and visually organized pages.

Finally, the talk will provide insights on future research directions in this evolving field of document AI and the journey so far.