Granted Patent B1

US 12,147,513 · App. 18/633,293 · Granted Nov 19, 2024

Inventors: Payal Jain (London, GB); Tariq Husayn Maonah (London, GB); Mariusz Saternus (Cracow, PL); Daniel Lewandowski (Cracow, PL); Biraj Krushna Rath (London, GB); Stuart Murray (London, GB); Philip Davies (London, GB)

Assignee: Citibank, N.A.

G06F21/31G06F21/6218G06F40/20

Abstract

Claims (112)

1. A non-transitory computer-readable storage medium comprising instructions thereon, wherein the instructions when executed by at least one data processor of a system, cause the system to:

receive an output generation request from a user device,

wherein the user device is associated with an authentication token, and

wherein the output generation request includes a prompt for generation of a text-based output using a first large-language model (LLM);

authenticate the user device based on the authentication token;

determine a performance metric value associated with the output generation request,

wherein the performance metric value indicates an estimated resource requirement for the output generation request;

identify, based on an attribute of the output generation request, a first prompt validation model of a plurality of prompt validation models;

provide the output generation request to the first prompt validation model to modify the prompt,

wherein modifying the prompt comprises:

determining that the prompt includes a forbidden token; and

generating the modified prompt by omitting the forbidden token;

compare the performance metric value of the output generation request with a first performance criterion associated with the first LLM of a plurality of LLMs;

in response to determining that the performance metric value satisfies the first performance criterion, provide the prompt to the first LLM to generate an output;

provide the output to an output validation model to generate a validation indicator associated with the output; and

in response to generating the validation indicator, transmit the output to a server system enabling access to the output by the user device.

2. The non-transitory computer-readable storage medium of claim 1 , wherein the instructions for authenticating the user device based on the authentication token cause the system to:

determine a user identifier associated with the user device;

determine, from a token database, a stored token associated with the user identifier;

compare the stored token and the authentication token associated with the output generation request; and

in response to determining that the stored token and the authentication token associated with the output generation request match, authenticate the user device.

3. The non-transitory computer-readable storage medium of claim 1 , wherein the instructions further cause the system to:

generate, based on the output generation request, an event record including (1) the performance metric value, (2) a user identifier associated with the user device, and (3) the prompt; and

transmit, to the server system, the event record for storage in an event database.

4. The non-transitory computer-readable storage medium of claim 1 , wherein the instructions for modifying the prompt cause the system to:

determine that the prompt includes a first alphanumeric token;

determine that one or more records in a sensitive token database include a representation of the first alphanumeric token; and

modify the prompt to include a second alphanumeric token in lieu of the first alphanumeric token,

wherein the sensitive token database does not include a record representing the second alphanumeric token.

5. The non-transitory computer-readable storage medium of claim 1 , wherein the instructions for modifying the prompt cause the system to:

generate a trace token comprising a traceable alphanumeric token; and

generate the modified prompt to include the trace token.

6. The non-transitory computer-readable storage medium of claim 1 , wherein the instructions for comparing the performance metric value with the first performance criterion cause the system to:

generate a cost metric value associated with the estimated resource requirement for the output generation request;

determine a threshold cost associated with the first LLM; and

determine that the cost metric value satisfies the threshold cost.

7. The non-transitory computer-readable storage medium of claim 1 , wherein the instructions for generating the validation indicator cause the system to:

extract a code sample from the output, wherein the code sample includes code for a software routine;

compile, within a virtual machine of the system, the code sample to generate an executable program associated with the software routine;

execute, within the virtual machine, the software routine using the executable program;

detect an anomaly in the execution of the software routine; and

in response to detecting the anomaly in the execution of the software routine, generate the validation indicator to include an indication of the anomaly.

8. A system comprising:

at least one hardware processor; and

at least one non-transitory memory storing instructions, which, when executed by the at least one hardware processor, cause the system to:

receive an output generation request from a user device,

wherein the user device is associated with an authentication token, and

wherein the output generation request includes a prompt for generation of a text-based output using a first large-language model (LLM);

authenticate the user device based on the authentication token;

determine a performance metric value associated with the output generation request,

wherein the performance metric value indicates an estimated resource requirement for the output generation request;

provide the output generation request to a first prompt validation model to modify the prompt,

wherein modifying the prompt comprises:

generating a trace token comprising a traceable alphanumeric token; and

generating the modified prompt to include the trace token;

compare the performance metric value of the output generation request with a first performance criterion associated with the first LLM of a plurality of LLMs;

in response to determining that the performance metric value satisfies the first performance criterion, provide the prompt to the first LLM to generate an output;

provide the output to an output validation model to generate a validation indicator associated with the output; and

in response to generating the validation indicator, transmit the output to a server system enabling access to the output by the user device.

9. The system of claim 8 , wherein the instructions for authenticating the user device based on the authentication token cause the system to:

determine a user identifier associated with the user device;

determine, from a token database, a stored token associated with the user identifier;

compare the stored token and the authentication token associated with the output generation request; and

in response to determining that the stored token and the authentication token associated with the output generation request match, authenticate the user device.

10. The system of claim 8 , wherein the instructions further cause the system to:

generate, based on the output generation request, an event record including (1) the performance metric value, (2) a user identifier associated with the user device, and (3) the prompt; and

transmit, to the server system, the event record for storage in an event database.

11. The system of claim 8 , wherein the instructions for modifying the prompt cause the system to:

determine that the prompt includes a first alphanumeric token;

determine that one or more records in a sensitive token database include a representation of the first alphanumeric token; and

modify the prompt to include a second alphanumeric token in lieu of the first alphanumeric token,

wherein the sensitive token database does not include a record representing the second alphanumeric token.

12. The system of claim 8 , wherein the instructions for modifying the prompt cause the system to:

determine that the prompt includes a forbidden token; and

generate the modified prompt by omitting the forbidden token.

13. The system of claim 8 , wherein the instructions for comparing the performance metric value with the first performance criterion cause the system to:

generate a cost metric value associated with the estimated resource requirement for the output generation request;

determine a threshold cost associated with the first LLM; and

determine that the cost metric value satisfies the threshold cost.

14. The system of claim 8 , wherein the instructions for providing the output generation request to the first prompt validation model cause the system to:

identify an attribute of the output generation request;

determine, based on the attribute, the first prompt validation model of a plurality of prompt validation models; and

provide the output generation request to the first prompt validation model.

15. A method comprising:

receiving an output generation request from a user device,

wherein the output generation request is associated with an authentication token, and

wherein the output generation request includes a prompt for generation of a text-based output using a first large-language model (LLM);

authenticating the user device based on the authentication token;

determining a performance metric value associated with the output generation request,

wherein the performance metric value indicates an estimated resource requirement for the output generation request;

identifying a first prompt validation model of a plurality of prompt validation models;

providing the output generation request to the first prompt validation model to modify the prompt,

wherein modifying the prompt comprises:

determining that the prompt includes a forbidden token; and

generating the modified prompt by omitting the forbidden token;

comparing the performance metric value of the output generation request with a first performance criterion associated with the first LLM of a plurality of LLMs;

in response to determining that the performance metric value satisfies the first performance criterion, providing the prompt to the first LLM to generate an output;

providing the output to an output validation model to generate a validation indicator associated with the output; and

in response to generating the validation indicator, transmitting the output to a server system enabling access to the output by the user device.

16. The method of claim 15 , wherein authenticating the user device based on the authentication token comprises:

determining a user identifier associated with the user device;

determining, from a token database, a stored token associated with the user identifier;

comparing the stored token and the authentication token associated with the output generation request; and

in response to determining that the stored token and the authentication token associated with the output generation request match, authenticating the user device.

17. The method of claim 15 , comprising:

generating, based on the output generation request, an event record including (1) the performance metric value, (2) a user identifier associated with the user device, and (3) the prompt; and

transmitting, to the server system, the event record for storage in an event database.

18. The method of claim 15 , comprising:

determining that the prompt includes a first alphanumeric token;

determining that one or more records in a sensitive token database include a representation of the first alphanumeric token; and

modifying the prompt to include a second alphanumeric token in lieu of the first alphanumeric token,

wherein the sensitive token database does not include a record representing the second alphanumeric token.

Assignments (1)

Cited By (11)

US 12,505,291 US 12,517,724 US 12,572,856 US 12,596,738 US 12,596,813 US 12,602,418 US 12,602,624 US 12,634,269 US 12,634,309 US 12,681,830 US 12,694,343