Granted Patent B1

US 12,450,442 · App. 19/057,835 · Granted Oct 21, 2025

Inventors: Rajkumar Janakiraman (Sammamish, WA); Ranjitha Gurunath Kulkarni (Sunnyvale, CA); James Johnson (San Marcos, CA)

Assignee: Dropbox, Inc.

G06F40/30G06F16/3329

Abstract

Claims (82)

1. A computer-implemented method comprising:

determining, utilizing a prompt classification model, that a prompt received from a client device involves one or more logical problems;

based on determining that the prompt involves the one or more logical problems, generating a logic code segment by processing the prompt utilizing a context engine comprising one or more large language models;

generating a logic result for the prompt by processing the logic code segment using a logic engine that solves the one or more logical problems involved with the prompt according to a structure of the logic code segment; and

generating a response to the prompt based at least in part on the logic result.

2. The computer-implemented method of claim 1 , further comprising:

receiving, from the logic engine, an error result for the prompt;

generating, based on the error result, a modified logic code segment by determining missing logic constraints in the logic code segment for solving the one or more logical problems involved with the prompt using the logic engine; and

generating an updated logic response for the prompt by processing the modified logic code segment using the logic engine.

3. The computer-implemented method of claim 1 , further comprising:

generating, utilizing the context engine, a logical plan partitioning the prompt into the one or more logical problems; and

based on the logical plan, generating a logic code segment that solves the one or more logical problems involved with the prompt.

4. The computer-implemented method of claim 1 , further comprising:

generating an additional logic result from the prompt by processing the logic code segment using an additional logic engine that solves the one or more logical problems involved with the prompt according to a different structure of the logic code segment.

5. The computer-implemented method of claim 1 , further comprising:

generating a first logic result utilizing a first logic engine to solve a first logical problem within the prompt;

generating a second logic result utilizing a second logic engine to solve a second logical problem within the prompt; and

generating the response to the prompt based at least in part on the first logic result and the second logic result.

6. The computer-implemented method of claim 1 , further comprising:

generating the response to the prompt with a retrieval augmented generation (RAG) system based at least in part on one or more source content items associated with a user account of a content management system; and

determining an accuracy of the response based on the logic result for the prompt.

7. The computer-implemented method of claim 1 , further comprising:

determining, utilizing a prompt classification model, that an additional prompt received from the client device involves one or more mathematical problems;

based on determining that the additional prompt involves the one or more mathematical problems, generating a mathematic code segment by processing the additional prompt utilizing the context engine comprising one or more large language models;

generating a mathematic result for the additional prompt by processing the mathematic code segment using a mathematic engine that solves the one or more mathematical problems involved with the additional prompt according to a structure of the mathematic code segment; and

generating a response to the additional prompt based at least in part on the mathematic result.

8. A system comprising:

at least one processor; and

a non-transitory computer readable medium comprising instructions that, when executed by the at least one processor, cause the system to:

determine, utilizing a prompt classification model, that a prompt received from a client device involves one or more logical problems;

based on determining that the prompt involves the one or more logical problems, generate one or more logic code segments by processing the prompt utilizing a context engine comprising one or more large language models;

generate one or more logic results for the prompt by processing the one or more logic code segments using a logic engine that solves the one or more logical problems involved with the prompt according to a structure of the one or more logic code segments;

generate a response to the prompt based at least in part on the one or more logic results; and

provide the response for display on a graphical user interface of the client device.

9. The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:

determine one or more missing logic constraints for solving the one or more logical problems involved with the prompt;

generate a modified logic code segment by adding the one or more missing logic constraints to a logic code segment to inform the logic engine; and

generate an updated logic response for the prompt by processing the modified logic code segment using the logic engine.

10. The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:

generate, utilizing the context engine, a logical plan partitioning the prompt into one or more ordered steps corresponding to the one or more logical problems; and

based on the logical plan, generate one or more logic results that solve the one or more logical problems involved with the prompt according to the one or more ordered steps.

11. The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:

identify, utilizing the context engine, a time phrase within the prompt;

generate, utilizing a large language model, a time code segment representing the time phrase;

generate a time block from the time code segment utilizing an interpreter engine coupled to the context engine;

generate a time result utilizing the context engine to process the time block; and

generate the response based at least in part on the time result.

12. The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:

generate a first logic result utilizing a first logic engine to solve a first logical problem within the prompt;

generate a second logic result utilizing a second logic engine to solve a second logical problem within the prompt that differs from the first logical problem; and

generate the response to the prompt based at least in part on the first logic result and the second logic result.

13. The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:

generate the response to the prompt with a retrieval augmented generation (RAG) system based at least in part on one or more source content items associated with a user account of a content management system;

generate a dependency graph reflecting relationships between the one or more source content items, the prompt, and the response; and

determine an accuracy of the response based on the dependency graph.

14. The system of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the system to:

determine that an additional prompt received from the client device involves one or more mathematical problems;

based on determining that the additional prompt involves the one or more mathematical problems, generate a mathematic code segment by processing the additional prompt utilizing the context engine;

generate a mathematic result for the additional prompt by processing the mathematic code segment using a mathematic engine that solves the one or more mathematical problems involved with the additional prompt; and

generate a response to the additional prompt based at least in part on the mathematic result.

15. A non-transitory computer readable medium comprising instructions that, when executed by at least one processor, cause the at least one processor to:

determine, utilizing a prompt classification model, that a prompt received from a client device involves one or more logical problems;

based on determining that the prompt involves the one or more logical problems, generate a logic code segment by processing the prompt utilizing a context engine comprising one or more large language models;

select a logic engine to solve the one or more logical problems involved with the prompt, wherein the selected logic engine solves the one or more logical problems according to a structure of the logic code segment;

generate a logic result for the prompt by processing the logic code segment using the selected logic engine; and

generate a response to the prompt based at least in part on the logic result.

16. The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:

receive, from the logic engine, an error result indicating one or more missing logic constraints for solving the one or more logical problems involved with the prompt;

based on the error result, request from the client device an additional prompt with information related to the one or more missing logic constraints;

based on receiving the additional prompt, generate a modified logic code segment by adding the one or more missing logic constraints to the logic code segment to inform the logic engine; and

generate an updated logic response for the prompt by processing the modified logic code segment using the logic engine.

17. The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:

receive, from the logic engine, an error result for the prompt indicating the one or more logical problems involved with the prompt are unsolvable; and

provide, for display on the client device, an error result notification comprising a reason why the one or more logical problems involved with the prompt are unsolvable.

18. The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:

generate, utilizing a large language model, a logical plan partitioning the prompt into one or more ordered steps for solving the one or more logical problems; and

based on the logical plan, generate a logic code segment that solves the one or more logical problems involved with the prompt.

19. The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:

generate an additional logic result from the prompt that solves a logical problem from the one or more logical problems by processing the logic code segment using an additional logic engine.

20. The non-transitory computer readable medium of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to:

determine one or more logical problem types for the one or more logical problems; and

select one or more logic engines based on the one or more logical problem types of the one or more logical problems.

Assignments (1)

References Cited (20)