Quickly add a zero to the bill with Photoshop, and the insurance company reimburses a multi-digit amount that was never charged. Didn't have your teeth cleaned this year? With a changed date, the health insurance company will certainly wave the same bill through this year as well. Nowadays, such forgeries are sometimes so good that a clerk cannot see them with the naked eye. The technology from natif.ai takes a closer look. Every pixel is checked. The company's proprietary document recognition technology makes this possible. Until now, natif.ai has supported its customers in classifying documents fully automatically and extracting all relevant data points. For this purpose, natif.ai offers a platform (https://platform.natif.ai) that allows developers to automate their document-based processes. With the support of the CISPA incubator, the startup company has now set itself the goal of detecting forgeries among the digitized documents processed using AI. An application for the StartUpSecure funding program was recently approved.
In the future, the AI will be able to detect the smallest deviations in font type, color, orientation, or size that occur as artifacts when a document is processed in Photoshop. It will be trained to recognize that something is wrong. Are the delivery date and invoice date too far apart? Does the item description match the price and the tax number match the vendor? "The AI knows where it's worth forging. It learns from the pixels and checks how plausible the content of the document is," explains Nico Herbig. The recognition is also very fast, he adds. Nico: "We process everything with deep learning approaches on graphics cards. The forgery detection will also run on graphics cards so that we can enable real-time use cases."
The StartUpSecure funding program offered by the German Federal Ministry of Education and Research and CISPA, which includes a research component, now offers natif.ai the opportunity to expand its business field and implement forgery detection alongside pure document extraction. Nico: "There are already approaches to forgery detection in academia, but unlike us, they usually work without document understanding." Incubation at CISPA brings both pieces of the puzzle together: With the know-how from CISPA research and the knowledge of the already existing solution, natif.ai can make document AI more secure against counterfeiting.
Dr. Nico Herbig earned his doctorate at the German Research Center for Artificial Intelligence (DFKI) and made the intersection between Natural Language Processing and Human Computer Interaction his topic. As a first hour employee at natif.ai, Nico handles product direction and development, as well as acquisition and execution of research projects.
