US 12,443,894 B1
Validating artificial intelligence model outputs using hash signatures and chunk-level access controls
Ganesh Prasad Bhat, New Jersey, NJ (US); Joshua Adam Goldman, Merrick, NY (US); Venkata Uttam Kumar Chunduri, New Jersey, NJ (US); Vishal Mysore, Mississauga (CA); Ramkumar Ayyadurai, Jersey City, NJ (US); and Chamindra Desilva, London (GB)
Assigned to CITIBANK, N.A.
Filed by Citibank, N.A., New York, NY (US)
Filed on May 14, 2025, as Appl. No. 19/207,355.
Application 19/015,646 is a division of application No. 18/653,858, filed on May 2, 2024, granted, now 12,198,030, issued on Jan. 14, 2025.
Application 19/207,355 is a continuation in part of application No. 19/015,646, filed on Jan. 10, 2025.
Application 18/653,858 is a continuation in part of application No. 18/637,362, filed on Apr. 16, 2024, granted, now 12,111,754, issued on Oct. 8, 2024.
Int. Cl. G06N 20/20 (2019.01)
CPC G06N 20/20 (2019.01) 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for generating responses using data retrieved by validating vector embeddings using hash signatures, the method comprising:
receiving, via a user interface of a computing device, an output generation request to generate an output using an input that includes (1) a plurality of documents and (2) a command set,
wherein each document of the plurality of documents includes a data chunk set comprising one or more of: text data, audio data, image data, or video data;
selecting a data chunk subset from respective data chunk sets of the plurality of documents by determining that each data chunk of the data chunk subset fails to map to one or more respective hash signatures produced by applying one or more hash functions on the data chunk;
validating each data chunk of the data chunk subset against a predefined constraint set defining an operative boundary set of a guideline set by:
identifying a content set within one or more data chunks of the data chunk subset failing to satisfy the predefined constraint set,
modifying the content set to satisfy the predefined constraint set by one or more of: (1) applying a data mask, (2) adding supplemental data, or (3) removing a portion of the content set, and
assigning a unique hash signature to the data chunk indicative of a degree of satisfaction of the data chunk with the guidelines;
generating, using an artificial intelligence (AI) model set, a response that is responsive to the input of the output generation request by:
transmitting the respective data chunk sets of the plurality of documents to one or more input nodes of the AI model set,
wherein each data chunk of the respective data chunk sets of the plurality of documents is associated with a corresponding unique hash signature, and
receiving, from one or more output nodes of the AI model set, the response;
responsive to generating the response, displaying a graphical layout on the user interface of the computing device that includes (1) a first graphical representation indicative of the input of the output generation request and (2) a second graphical representation indicative of the response.