IP Library › Granted Patent US 12,265,570
Granted Patent B2
US 12,265,570 · App. 18/542,481 · Granted Apr 1, 2025

Generative artificial intelligence enterprise search

Inventors: Thomas M. Siebel (Woodside, CA); Nikhil Krishnan (Los Altos, CA); Louis Poirier (Paris, FR); Michael Haines (San Francisco, CA); Romain Juban (San Francisco, CA)
Assignee: C3.ai, Inc.
G06F16/345G06F16/3326G06F16/334G06F16/3347G06F16/335G06F16/338G06F40/20G06F40/40G06N3/092G06N5/04G06N20/00
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,265,570
App. No.
18/542,481
Filed
Dec 15, 2023
Granted
Apr 1, 2025
Kind
B2
Art Unit
2168
USPC
707/728
Abstract

Systems and methods are configured to generate a set of potential responses to a prompt using one or more data models with data from at least a plurality of data domains of an enterprise information environment that includes access controls. A deterministic response is selected from the set of potential responses based on scoring of the validation data and restricting based on access controls in view of profile information associated with the prompt. These enterprise generative AI systems and methods support granular enterprise access controls, privacy, and security requirements, and provide traceable references and links to source information underlying the generative AI insights. These systems and methods enable dramatically increased utility for enterprise users to access information, analyses, and predictive analytics associated with and derived from a combination of enterprise and external information systems.

Claims (48)

1. A method comprising:

generating a set of potential responses to a prompt using one or more data models with data from at least a plurality of data domains of an enterprise information environment that includes access controls, wherein generating the set of potential responses uses at least one large language model trained on the at least the plurality of data domains of the enterprise information environment to determine data object semantics, and wherein the at least one large language model includes vectorized data from the data from the at least the plurality of data domains with embeddings;

determining validation data for the set of potential responses, wherein the validation data is from the at least the plurality of data domains of the enterprise information environment, wherein determining the validation data comprises retrieving the embeddings from the vectorized data;

selecting a deterministic response from the set of potential responses based on scoring of the validation data and restricting the deterministic response based on the access controls in view of profile information associated with the prompt, wherein selecting the deterministic response from the set of potential responses comprises using the embeddings from the vectorized data to determine relevance evaluations; and

outputting the selected deterministic response with the validation data corresponding to the selected deterministic response.

2. The method of claim 1 , further comprising:

generating a traceability analysis of the validation data, the traceability analysis indicating any of documents, document segments, and insights of at least a portion of one or more enterprise data sets.

3. The method of claim 1 , wherein the scoring of the validation data comprises:

determining, based on the one or more data models, a plurality of relevance scores associated with at least a portion of each piece of validation data for the set of potential responses.

4. The method of claim 1 ,

wherein at least two of the data domains include industry-specific data for different industries, wherein at least one response of the set of potential responses uses a first data domain of the at least the plurality of data domains, and wherein at least another response of the set of potential responses uses a second data domain of the at least the plurality of data domains,

wherein the one or more data models comprises multiple models trained for different data domains with industry-specific data of the at least the plurality of data domains,

wherein each data model represents respective relationships and attributes of the corresponding different data domain industry-specific data of the different data domains, and

wherein the respective relationships and attributes include any of data types, data formats, and industry-specific information.

5. The method of claim 1 , wherein the one or more data models include multimodal models wherein at least one is a large language model.

6. The method of claim 1 , wherein the access controls enforce restrictions including at least one of administrative policies, security policies, profile rights, and organizational controls.

7. The method of claim 1 , wherein the access controls cause a different deterministic response to be selected based on profiles with different access rights.

8. The method of claim 1 , wherein the access controls cause a different validation data to be output based on profiles with different access rights.

9. The method of claim 1 , wherein the selected deterministic response comprises at least one of predictions, insights, or recommendations from an artificial intelligence application.

10. The method of claim 1 , wherein the output includes at least one of data visualization, automated control and instruction, a report, and a dynamically configured dashboard.

11. The method of claim 1 , wherein generating the set of potential responses using the one or more data models comprises using a retrieval model to determine relevant objects from the at least the plurality of data domains of the enterprise information environment.

12. The method of claim 1 , wherein the data object semantics are for enterprise concepts, features, and components.

13. The method of claim 1 ,

wherein the one or more data models uses a model driven architecture with a type system for disparate data formats of the at least the plurality of data domains, and

wherein determining the validation data uses the type system of the model driven architecture for access controls of the validation data in view of the profile information associated with the prompt.

14. The method of claim 1 , wherein outputting the selected deterministic response includes a generative AI insight and a traceability report indicating at least one source document used to determine the selected deterministic response, and wherein the at least one source document is from a particular data domain of the at least the plurality of data domains of the enterprise information environment.

15. A system comprising:

one or more processors; and

memory storing instructions that, when executed by the one or more processors, cause the system to perform:

generating a set of potential responses to a prompt using one or more data models with data from at least a plurality of data domains of an enterprise information environment that includes access controls, wherein generating the set of potential responses uses at least one large language model trained on the at least the plurality of data domains of the enterprise information environment to determine data object semantics, and wherein the at least one large language model includes vectorized data from the data from the at least the plurality of data domains with embeddings;

determining validation data for the set of potential responses, wherein the validation data is from the at least the plurality of data domains of the enterprise information environment, wherein determining the validation data comprises retrieving the embeddings from the vectorized data;

selecting a deterministic response from the set of potential responses based on scoring of the validation data and restricting the deterministic response based on the access controls in view of profile information associated with the prompt; wherein selecting the deterministic response from the set of potential responses comprises using the embeddings from the vectorized data to determine relevance evaluations; and

outputting the selected deterministic response with the validation data corresponding to the selected deterministic response.

16. The system of claim 15 , further comprising:

generating a traceability analysis of the validation data, the traceability analysis indicating any of documents, document segments, and insights of at least a portion of one or more enterprise data sets.

17. The system of claim 15 , wherein the scoring of the validation data comprises:

determining, based on the one or more data models, a plurality of relevance scores associated with at least a portion of each piece of validation data for the set of potential responses.

18. The system of claim 15 , wherein the one or more data models include multimodal models wherein at least one is a large language model.

19. The system of claim 15 , wherein the access controls enforce restrictions including at least one of administrative policies, security policies, profile rights, and organizational controls.

20. The system of claim 15 , wherein the access controls cause a different deterministic response to be selected based on profiles with different access rights.

21. The system of claim 15 , wherein the access controls cause a different validation data to be output based on profiles with different access rights.

22. The system of claim 15 , wherein the selected deterministic response comprises at least one of predictions, insights, or recommendations from an artificial intelligence application.

23. The system of claim 15 , wherein the output includes at least one of data visualization, automated control and instruction, a report, and a dynamically configured dashboard.

24. A non-transitory computer readable medium comprising instructions that, when executed, cause one or more processors to perform:

generating a set of potential responses to a prompt using one or more data models with data from at least a plurality of data domains of an enterprise information environment that includes access controls, wherein generating the set of potential responses uses at least one large language model trained on the at least the plurality of data domains of the enterprise information environment to determine data object semantics, and wherein the at least one large language model includes vectorized data from the data from the at least the plurality of data domains with embeddings;

determining validation data for the set of potential responses, wherein the validation data is from the at least the plurality of data domains of the enterprise information environment, wherein determining the validation data comprises retrieving the embeddings from the vectorized data;

selecting a deterministic response from the set of potential responses based on scoring of the validation data and restricting the deterministic response based on the access controls in view of profile information associated with the prompt, wherein selecting the deterministic response from the set of potential responses comprises using the embeddings from the vectorized data to determine relevance evaluations; and

outputting the selected deterministic response with the validation data corresponding to the selected deterministic response.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Mar 7, 2024
From: SIEBEL, THOMAS M.; KRISHNAN, NIKHIL; POIRIER, LOUIS; HAINES, MICHAEL; JUBAN, ROMAIN
To: C3.AI, INC.
Reel/Frame 066689/0436 →
Continuity (4)
Provisional Application 63492133 · Mar 24, 2023
Provisional Application 63446792 · Feb 17, 2023
Provisional Application 63433124 · Dec 16, 2022
Related Publication 20240202221A1 · Jun 20, 2024
References Cited (64)
US 8666928B2 · Tunstall-Pedoe · 2014 [cited by applicant]
US 10467347B1 · Reiter · 2019 [cited by applicant]
US 10621282B1 · Selfridge · 2020 [cited by applicant]
US 10827071B1 · Adibi · 2020 [cited by applicant]
US 10943072B1 · Jaganmohan · 2021 [cited by applicant]
US 11138473B1 · Padmanabhan · 2021 [cited by applicant]
US 11169798B1 · Palmer · 2021 [cited by applicant]
US 11302310B1 · Gandhe · 2022 [cited by examiner]
US 11516158B1 · Luzhnica · 2022 [cited by applicant]
US 11783805B1 · Nadig · 2023 [cited by applicant]
US 11900289B1 · Kumar · 2024 [cited by applicant]
US 11908476B1 · Lyu · 2024 [cited by applicant]
US 20010053968A1 · Galitsky · 2001 [cited by applicant]
US 20020099658A1 · Nielsen · 2002 [cited by applicant]
US 20050154580A1 · Horowitz · 2005 [cited by applicant]
US 20060294002A1 · Brett · 2006 [cited by applicant]
US 20070055656A1 · Tunstall-Pedoe · 2007 [cited by applicant]
US 20090292687A1 · Fan · 2009 [cited by applicant]
US 20100332585A1 · Driesen · 2010 [cited by applicant]
US 20110307435A1 · Overell · 2011 [cited by applicant]
US 20140272884A1 · Allen · 2014 [cited by examiner]
US 20170148434A1 · Monceaux · 2017 [cited by applicant]
US 20170293666A1 · Ragavan · 2017 [cited by examiner]
US 20170293864A1 · Oh · 2017 [cited by applicant]
US 20180081934A1 · Byron · 2018 [cited by examiner]
US 20180191867A1 · Siebel · 2018 [cited by examiner]
US 20180246983A1 · Rathod · 2018 [cited by applicant]
US 20180314762A1 · Rathod · 2018 [cited by applicant]
US 20180330258A1 · Harris · 2018 [cited by examiner]
US 20180367483A1 · Rodriguez · 2018 [cited by applicant]
US 20190034813A1 · Das · 2019 [cited by examiner]
US 20190121612A1 · Ashoori · 2019 [cited by applicant]
US 20190132203A1 · Wince · 2019 [cited by applicant]
US 20190179940A1 · Ross · 2019 [cited by applicant]
US 20190251169A1 · Loghmani · 2019 [cited by applicant]
US 20190266157A1 · Brown · 2019 [cited by applicant]
US 20190272323A1 · Galitsky · 2019 [cited by applicant]
US 20190361961A1 · Zambre · 2019 [cited by applicant]
US 20200012584A1 · Walters · 2020 [cited by applicant]
US 20200210647A1 · Panuganty · 2020 [cited by applicant]
US 20210142177A1 · Mallya · 2021 [cited by applicant]
US 20210158897A1 · Chatzou · 2021 [cited by applicant]
US 20210173744A1 · Agrawal · 2021 [cited by applicant]
US 20220027578A1 · Chorakhalikar · 2022 [cited by applicant]
US 20220036153A1 · O'Malia · 2022 [cited by applicant]
US 20220066914A1 · Drain · 2022 [cited by applicant]
US 20220164109A1 · Hankins · 2022 [cited by applicant]
US 20220237368A1 · Tran · 2022 [cited by applicant]
US 20220261817A1 · Ferrucci · 2022 [cited by applicant]
US 20220339781A1 · Weng · 2022 [cited by applicant]
US 20220400094A1 · Sampath · 2022 [cited by applicant]
US 20230131495A1 · Tater · 2023 [cited by applicant]
US 20230177878A1 · Sekar · 2023 [cited by examiner]
US 20240112394A1 · Neal · 2024 [cited by applicant]
US 20240135199A1 · Portisch · 2024 [cited by applicant]
US 20240320867A1 · Bean · 2024 [cited by applicant]
Contextual AI, Inc., “Introducing RAG 2.0,” Mar. 19, 2024 [retrieved online at contextual.ai/introducing-rag2]. [cited by applicant]
Christidis, Angelos et al., “Serving Machine Workloads in Resource Constrained Environments: A Serverless Deployment Example,” 2019 IEEE 12th Conference on Service-Oriented Computing and Applications (SOCA), pp. 55-63, … [cited by applicant]
Lewis, Patrick et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” Proceedings of the 34th International Conference on Neural Information Processing Systems, Dec. 6, 2020. [cited by applicant]
International Application No. PCT/US2023/084462, Search Report and Written Opinion dated Apr. 1, 2024. [cited by applicant]
International Application No. PCT/US2023/084465, Search Report and Written Opinion dated Apr. 11, 2024. [cited by applicant]
International Application No. PCT/US2023/084468, Search Report and Written Opinion dated Apr. 16, 2024. [cited by applicant]
International Application No. PCT/US2023/084481, Search Report and Written Opinion dated Mar. 25, 2024. [cited by applicant]
International Application No. PCT/US2023/084456, Search Report and Written Opinion dated Apr. 12, 2024. [cited by applicant]
Cited By (4)
US 12,430,333 US 12,511,282 US 12,614,080 US 12,657,262