US 11,874,823 B1
Agnostic image digitizer to detect fraud
James Siekman, Charlotte, NC (US); Aubrey Breon Farrar, Sr., Waldorf, MD (US); Mohamed Faris Khaleeli, Charlotte, NC (US); Patricia Ann Albritton, Charlotte, NC (US); Sheila Page, Charlotte, NC (US); Mark Alan Odiorne, Waxhaw, NC (US); and Marcus R. Matos, Richardson, TX (US)
Assigned to Bank of America Corporation, Charlotte, NC (US)
Filed by Bank of America Corporation, Charlotte, NC (US)
Filed on Jul. 27, 2022, as Appl. No. 17/874,714.
Int. Cl. G06F 16/00 (2019.01); G06F 16/23 (2019.01); G06F 21/62 (2013.01)
CPC G06F 16/2365 (2019.01) [G06F 21/6218 (2013.01)] 3 Claims
OG exemplary drawing
 
1. A method for extending a usable lifetime of a legacy database of an organization by converting documents stored on the legacy database from a format that is incompatible with a computer hardware processor in electronic communication with the legacy database into a format compatible with the computer hardware processor and using the converted documents to train a machine learning artificial intelligence (AI) system to auto-populate a newly requested document, the method comprising:
digitizing, using optical character recognition (OCR) run on the computer hardware processor, documents in a first format into a digital format;
wherein:
the first format is incompatible with the computer hardware processor; and
the documents are stored on the legacy database;
converting, using the computer hardware processor, the documents in the digital format into a second format;
wherein the second format is compatible with the computer hardware processor;
storing, using the computer hardware processor, the documents in the second format on the legacy database;
training, using a graphics processing unit (GPU) in electronic communication with the computer hardware processor, a machine learning AI system, using the documents stored in the second format in the legacy database;
wherein the machine learning AI system auto-populates new documents in the second format;
receiving a request, at the computer hardware processor, from an entity for a first document in the second format;
determining a confidence level, using the GPU, for completing the first document in the second format;
when the confidence level is above a pre-determined threshold, using the GPU to run the machine learning AI system to auto-populate the first document in the second format;
providing, using the computer hardware processor, the auto-populated first document in the second format to the entity for feedback;
receiving from the entity, at the computer hardware processor, the first document in corrected form indicating that there was a mistake in the auto-population of the first document;
storing, using the computer hardware processor, the first document in corrected form on the legacy database;
when the confidence level falls below the pre-determined threshold due to the mistake in the auto-population of the first document, updating the training of the machine learning AI system, using the GPU, to learn from the mistake;
receiving a request, at the computer hardware processor, from an entity for a second document in the second format;
determining, using the GPU, a confidence level for completing the second document;
when the confidence level for completing the second document is above the pre-determined threshold, using the GPU to run the machine learning AI system to auto-populate the second document;
providing, using the computer hardware processor, the auto-populated second document to the entity for feedback;
receiving, at the computer hardware processor, feedback from the entity that the second document is auto-populated correctly;
storing, using the computer hardware processor, the second document on the legacy database; and
updating, using the GPU, the machine learning AI system using the feedback that the second document is auto-populated correctly.