Granted Patent B1

US 12,367,425 · App. 19/005,699 · Granted Jul 22, 2025

Inventors: Elaine Kelsey (Corvallis, OR); Sazzad Mahmud Nasir (Muncie, IN); Jeffrey Thomas Yarbro (Memphis, TN); Lauren Elizabeth Egerton (New York, NY); Elliot Nicholas Robson (Seoul, KR); Brendan Michael Kelly (Somerville, MA); Robert Oscar Robson (Corvallis, OR); Spencer Thomas Ward (Kent, WA)

Assignee: THIA ST CO.

G06N20/00

Abstract

Claims (139)

1. A method, comprising:

(a) capturing copilot objectives;

(b) obtaining one or more document records, each document record pertaining to one or more of the copilot objectives;

(c) identifying one or more data sources supporting the document record(s);

(d) configuring one or more data producers to provide one or more interfaces with the data source(s);

for each of the data producer(s):

(e) executing a respective first training regimen on the each data producer;

for each of a plurality of microservices of a copilot flow:

(f) executing a respective second training regimen on the each microservice;

wherein executing the first or second training regimen on a target comprises:

(ef1) generating training records including disjoint first and second subsets of records;

(ef2) applying the first subset of records to train the target; and

(ef3) applying the second subset of records to test the target; and

further wherein:

act (ef1) comprises act (ef1a) or act (ef1b):

(ef1a) by an expert, generating a first group of the training records; or

(ef1b) by a person distinct from the expert, generating first candidate training records and, by the expert, validating a subset of the first candidate training records as a second group of the training records; and

act (ef1) further comprises act (ef1c):

(ef1c) by a trained machine-learning tool, generating second candidate training records and, by the expert, validating a subset of the second candidate training records as a third group of the training records;

(g) assemble a copilot incorporating the microservices and the data producer(s):

(h) test the copilot; and

in at least a first case, in which the copilot satisfies a predetermined performance criterion:

(i) deploying the copilot.

2. The method of claim 1 , wherein act (h) determines that the copilot does not satisfy the predetermined performance criterion, and the method further comprises:

until the copilot satisfies the predetermined performance criterion:

performing one or more additional iterations of at least one among acts (a)-(g); and

repeating act (h).

3. The method of claim 1 , wherein the deployed copilot is configured to:

receive input from a client; and

process the input, using at least some of the microservices and the data producer(s), to obtain a result; and

transmit the result to the client.

4. The method of claim 1 , further comprising:

(i) configuring at least one of the identified data source(s) into a data repository;

wherein the assembled copilot further incorporates the data repository.

5. The method of claim 1 , wherein the document record(s) comprise one or more of:

a meeting transcript; an email conversation; or a presentation.

6. The method of claim 1 , wherein act (b) comprises:

prompting a trained machine learning tool to identify one or more candidate document records pertinent to one or more of the copilot objectives;

receiving the candidate document record(s) from the trained machine learning tool; and

validating a subset of the candidate document record(s) as at least some of the obtained document record(s).

7. The method of claim 1 , wherein act (c) comprises:

prompting a trained machine learning tool to determine one or more candidate data sources supporting one or more of the document record(s);

receiving the candidate data source(s) from the trained machine learning tool; and

validating a subset of the candidate data source(s) as at least some of the identified data source(s).

8. One or more computer-readable media storing instructions which, when executed by one or more hardware processors, cause the one or more hardware processors to perform operations comprising:

(a) receiving one or more task inputs within scope of one or more copilot objectives;

(b) obtaining one or more document records relevant to the task input(s), the obtaining including:

(b1) determining one or more candidate document records relevant to the task input(s); and

(b2) obtaining validation of a subset of the candidate document record(s);

wherein the obtained document record(s) comprise the validated subset of the candidate document record(s);

(c) based at least partly on the document record(s), identifying one or more data sources supporting the document record(s);

(d) linking one or more of the data source(s) to respective data producer(s) of a copilot instance;

for each of the data producer(s):

(e) applying respective training data to fine-tune the each data producer until a respective first performance criterion is satisfied; and

for each of a plurality of microservices of the copilot instance:

(f) applying respective training data to fine-tune the each microservice until a respective second performance criterion is satisfied; and

(g) apply testing data to verify that performance of the copilot instance, incorporating the fine-tuned data producer(s) and the fine-tuned microservices, satisfies a third performance criterion;

wherein the copilot instance is configured to receive new task inputs within scope of the copilot objective(s) and, in response, generate and deliver corresponding outputs satisfying the copilot objective(s).

9. The one or more computer-readable media of claim 8 , wherein the linked one or more data sources are one or more first data sources, and the operations further comprise:

(h) subsequent to act (c) and prior to act (g), integrating one or more second data sources of the one or more data sources into a data repository coupled to a retrieval microservice of the copilot instance.

10. The one or more computer-readable media of claim 8 , wherein the operations further comprise:

(i) integrating at least one of the fine-tuned data producer(s) into the copilot instance.

11. The one or more computer-readable media of claim 8 , wherein the new task inputs direct the copilot instance to conduct or monitor interviews.

12. The one or more computer-readable media of claim 8 , wherein the new task inputs direct the copilot instance to annotate an interview or annotate a recorded work session.

13. A method, comprising:

(a) capturing copilot objectives;

(b) obtaining one or more document records, each document record pertaining to one or more of the copilot objectives;

(c) identifying one or more data sources supporting the document record(s);

(d) configuring one or more data producers to provide one or more interfaces with the data source(s);

for each of the data producer(s):

(e) executing a respective first training regimen on the each data producer;

for each of a plurality of microservices of a copilot flow:

(f) executing a respective second training regimen on the each microservice;

wherein executing the first or second training regimen on a target comprises:

(ef1) generating training records including disjoint first and second subsets of records;

(ef2) applying the first subset of records to train the target; and

(ef3) applying the second subset of records to test the target;

wherein, for at least a first case for which the testing at act (ef3) determines that neither a predetermined second performance criterion nor a third criterion is met, the method further comprises:

(ef4) iterating acts (ef2)-(ef3) until the testing at act (ef3) determines that the predetermined second performance criterion or the third criterion is met;

(g) assemble a copilot incorporating the microservices and the data producer(s):

(h) test the copilot; and

in at least a first case, in which the copilot satisfies a predetermined performance criterion:

(i) deploying the copilot.

14. A method, comprising:

(a) capturing copilot objectives;

(b) obtaining one or more document records, each document record pertaining to one or more of the copilot objectives;

(c) identifying one or more data sources supporting the document record(s);

(d) configuring one or more data producers to provide one or more interfaces with the data source(s);

for each of the data producer(s):

(e) executing a respective first training regimen on the each data producer;

for each of a plurality of microservices of a copilot flow:

(f) executing a respective second training regimen on the each microservice;

wherein executing the first or second training regimen on a target comprises:

(ef1) generating training records including disjoint first and second subsets of records;

(ef2) applying the first subset of records to train the target; and

(ef3) applying the second subset of records to test the target;

wherein, for at least a first case for which the testing at act (ef3) determines that a predetermined second performance criterion is not met, the method further comprises iterating, until act (ef6) establishes that the predetermined second performance criterion is met:

(ef4) generating additional training records including disjoint first and second additional subsets of records;

(ef5) applying the first additional subset of records to train the target; and

(ef6) applying at least the second additional subset of records to test the target;

(g) assemble a copilot incorporating the microservices and the data producer(s):

(h) test the copilot; and

in at least a first case, in which the copilot satisfies a predetermined performance criterion:

(i) deploying the copilot.

15. A method, comprising:

(a) capturing copilot objectives;

(b) obtaining one or more document records, each document record pertaining to one or more of the copilot objectives;

(c) identifying one or more data sources supporting the document record(s);

(d) configuring one or more data producers to provide one or more interfaces with the data source(s);

for each of the data producer(s):

(e) executing a respective first training regimen on the each data producer;

for each of a plurality of microservices of a copilot flow:

(f) executing a respective second training regimen on the each microservice;

wherein the microservices include an expansion microservice, a retrieval microservice, a qualification microservice, a core microservice, a protection microservice, and an evaluation microservice;

(g) assemble a copilot incorporating the microservices and the data producer(s), by:

coupling the expansion microservice to receive input from a client interface or an intermodal microservice;

coupling the retrieval microservice to receive input from the expansion microservice, the data producer(s), and a data repository;

coupling the qualification microservice to receive input from the retrieval microservice;

coupling the core microservice to receive input from the qualification microservice;

coupling the protection microservice to receive input from the core microservice;

coupling the evaluation microservice to receive input from the protection microservice; and

coupling the evaluation microservice to provide output to the client interface or the retrieval microservice;

(h) test the copilot; and

in at least a first case, in which the copilot satisfies a predetermined performance criterion:

(i) deploying the copilot.

16. One or more computer-readable media storing instructions which, when executed by one or more hardware processors, cause the one or more hardware processors to perform operations comprising:

(a) receiving one or more task inputs within scope of one or more copilot objectives;

(b) obtaining one or more document records relevant to the task input(s);

(c) based at least partly on the document record(s), identifying one or more data sources supporting the document record(s), the identifying including:

(c1) determining one or more candidate data sources relevant to the document record(s); and

(c2) obtaining validation of a subset of the candidate data source(s);

wherein the identified data source(s) comprise the validated subset of the candidate data source(s);

(d) linking one or more of the data source(s) to respective data producer(s) of a copilot instance;

for each of the data producer(s):

(e) applying respective training data to fine-tune the each data producer until a respective first performance criterion is satisfied; and

for each of a plurality of microservices of the copilot instance:

(f) applying respective training data to fine-tune the each microservice until a respective second performance criterion is satisfied; and

(g) apply testing data to verify that performance of the copilot instance, incorporating the fine-tuned data producer(s) and the fine-tuned microservices, satisfies a third performance criterion;

wherein the copilot instance is configured to receive new task inputs within scope of the copilot objective(s) and, in response, generate and deliver corresponding outputs satisfying the copilot objective(s).

Assignments (1)

Continuity (7)

References Cited (70)