Granted Patent B2

US 12,399,907 · App. 18/898,506 · Granted Aug 26, 2025

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

Assignee: THIA ST Co.

G06F16/3329G06F40/284

Abstract

Claims (140)

1. A computer-implemented method for augmenting input for a core microservice of a copilot, comprising:

receiving second input, which reflects first input from a client;

recursively performing a plurality of iterations of retrieval augmented generation (RAG), each of the iterations comprising:

transmitting third output to a data producer; and

responsive to the transmitting, receiving fourth input from the data producer;

wherein, on an initial iteration of the iterations, the third output is based on the second input; and

wherein, on subsequent iterations of the iterations, the third output is based on the fourth input of an immediately preceding iteration of the iterations;

wherein the iterations are terminated upon meeting a stopping criterion comprising: two successive iterations of the iterations returning respective fourth inputs whose similarity is greater than or equal to a predetermined threshold; and

transmitting, toward the core microservice, fourth output comprising at least part of the fourth input received on a final iteration of the iterations;

wherein the fourth output enables the core microservice to generate a response to the first input from the client.

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

receiving the first input from the client;

extracting one or more first tokens from the first input;

determining one or more second tokens associated with but distinct from the first tokens; and

generating a first output combining the first and second tokens; and

transmitting the first output to a retrieval microservice, where the first output is received as the second input.

3. The computer-implemented method of claim 1 , wherein the second input comprises a plurality of tokens, and the method further comprises:

executing the recursively performing acts individually for each of the plurality of tokens to obtain respective final fourth inputs on respective final iterations;

wherein the fourth output comprises, for each of the plurality of tokens, the fourth input of a final iteration of the iterations.

4. The computer-implemented method of claim 1 , wherein the fourth output comprises the fourth input of each of the plurality of iterations.

5. The computer-implemented method of claim 1 , wherein one or more of the third output, the fourth input, or the fourth output comprises an array of elements, each of the elements comprising a document, a vector representation of a document, or a token derived from the second input.

6. One or more computer-readable media storing instructions which, when executed by one or hardware processors, cause the one or more hardware processors to perform operations for augmenting input for a core microservice of a copilot, the operations comprising:

receiving second input, which reflects first input from a client application;

recursively performing a plurality of iterations of retrieval augmented generation (RAG), each of the iterations comprising:

transmitting third output to a data producer; and

responsive to the transmitting, receiving fourth input from the data producer;

wherein, on an initial iteration of the iterations, the third output is based on the second input; and

wherein, on subsequent iterations of the iterations, the third output is based on the fourth input of an immediately preceding iteration of the iterations;

scoring, for relevance to the first or second input, constituents of the fourth inputs received on respective iterations of the plurality of iterations;

transmitting, toward the core microservice, fourth output comprising the fourth input received on a final iteration of the iterations;

retaining, in the fourth output, high-scoring constituents among the constituents of the fourth inputs; and

discarding low-scoring constituents among the constituents of the fourth inputs;

wherein the fourth output enables the core microservice to generate a response to the first input from the client application.

7. A system comprising:

one or more hardware processors, with memory coupled thereto; and

one or more computer readable media storing instructions comprising a plurality of modules which, when executed by the one or more hardware processors, implement respective microservices, the microservices forming a weakly connected network of microservices configured as a copilot for one or more first client applications;

wherein each of the microservices is configured to:

receive input from (i) a respective first group comprising one or more others of the microservices or (ii) one or more second client applications; and

transmit output to (i) a second group comprising one or more of the microservices or (ii) one or more third client applications;

wherein a plurality of the microservices incorporate respective trained machine learning tools;

wherein the network of microservices comprises at least a retrieval microservice, one or more data producers, a qualification microservice, and a core microservice;

wherein the retrieval microservice is configured to:

receive second input, which reflects first input from a given one of the one or more first client applications;

recursively perform a plurality of iterations of retrieval augmented generation (RAG), each of the iterations comprising:

transmitting third output to a respective data producer of the one or more data producers; and

responsive to the transmitting, receiving fourth input from the respective data producer;

wherein, on an initial iteration of the iterations, the third output is based on the second input; and

wherein, on subsequent iterations of the iterations, the third output is based on the fourth input of an immediately preceding iteration of the iterations; and

transmit, toward the core microservice, fourth output comprising the fourth input received on a final iteration of the iterations;

wherein the fourth output enables the core microservice to generate a response to the first input from the given first client application; and

wherein the qualification microservice configured to:

receive a ninth input based on the fourth output;

compare the ninth input with a graphical model of a knowledge corpus incorporated in the copilot;

determine whether the copilot is competent to act on the ninth input;

upon determining that the copilot is competent:

determine and transmit ninth output based on the ninth input, toward the core microservice; and

upon determining that the copilot is not competent, transmit a notification indicating lack of competence.

8. The system of claim 7 , wherein the respective data producer, at a given iteration of the iterations, comprises:

a document repository storing a plurality of documents; and

an index storing vector representations of the stored documents.

9. The system of claim 7 wherein, for all of the iterations, the respective data producer is a common data producer.

10. The system of claim 7 wherein, for given first and second iterations of the iterations, the respective data producer for the first iteration is distinct from the respective data producer for the second iteration.

11. A computer-implemented method for augmenting input for a core microservice of a copilot, comprising:

receiving second input, which reflects first input from a client;

recursively performing a plurality of iterations of retrieval augmented generation (RAG), each of the iterations comprising:

transmitting third output to a data producer;

responsive to the transmitting, receiving fourth input from the data producer;

scoring, for relevance to the first or second input, constituents of the fourth input;

retaining high-scoring constituents among the constituents of the fourth input; and

discarding low-scoring constituents among the constituents of the fourth input;

wherein, on an initial iteration of the iterations, the third output is based on the second input; and

wherein, on subsequent iterations of the iterations, the third output is based on the fourth input of an immediately preceding iteration of the iterations;

transmitting, toward the core microservice, fourth output comprising the retained constituents of the fourth input received on a final iteration of the iterations;

wherein the fourth output enables the core microservice to generate a response to the first input from the client.

12. A computer-implemented method for augmenting input for a core microservice of a copilot, comprising:

receiving second input, which reflects first input from a client;

recursively performing a plurality of iterations of retrieval augmented generation (RAG), each of the iterations comprising:

transmitting third output to a data producer; and

responsive to the transmitting, receiving fourth input from the data producer;

wherein, on an initial iteration of the iterations, the third output is based on the second input; and

wherein, on subsequent iterations of the iterations, the third output is based on the fourth input of an immediately preceding iteration of the iterations;

wherein the iterations are terminated upon meeting a stopping criterion comprising: an increase in volume over successive iterations being less than or equal to a predetermined threshold, the volume on each of the successive iterations being defined by the second input and the fourth input on the respective iteration, in a graphical representation of a knowledge domain; and

transmitting, toward the core microservice, fourth output comprising at least part of the fourth input received on a final iteration of the iterations;

wherein the fourth output enables the core microservice to generate a response to the first input from the client.

13. One or more computer-readable media storing instructions which, when executed by one or hardware processors, cause the one or more hardware processors to perform first operations for augmenting input for a core microservice of a copilot and second operations of a data producer, the first operations comprising:

receiving second input, which reflects first input from a client application;

recursively performing a plurality of iterations of retrieval augmented generation (RAG), each of the iterations comprising:

transmitting third output to the data producer; and

responsive to the transmitting, receiving fourth input from the data producer;

wherein, on an initial iteration of the iterations, the third output is based on the second input; and

wherein, on subsequent iterations of the iterations, the third output is based on the fourth input of an immediately preceding iteration of the iterations; and

transmitting, toward the core microservice, fourth output comprising the fourth input received on a final iteration of the iterations;

wherein the fourth output enables the core microservice to generate a response to the first input from the client application; and

the second operations comprising:

receiving, on a given iteration of the plurality of iterations, a seventh input based on the third output;

retrieving, from one or more databases, database objects relevant to the seventh input; and

determining and transmitting a seventh output based on the retrieved database objects;

wherein, on the given iteration, the fourth input corresponding to the third output is based on the seventh output.

14. The one or more computer-readable media of claim 13 , wherein the iterations are terminated upon meeting a stopping criterion, wherein:

the stopping criterion comprises completion of a predetermined number of the iterations.

15. The one or more computer-readable media of claim 13 , wherein the iterations are terminated upon meeting a stopping criterion, wherein:

the stopping criterion comprises an amount of new data in the fourth input being less than or equal to a first predetermined threshold.

16. One or more computer-readable media storing instructions which, when executed by one or hardware processors, cause the one or more hardware processors to perform first operations for augmenting input for a core microservice of a copilot and second operations of a qualification microservice, the first operations comprising:

receiving second input, which reflects first input from a client application;

recursively performing a plurality of iterations of retrieval augmented generation (RAG), each of the iterations comprising:

transmitting third output to a data producer; and

responsive to the transmitting, receiving fourth input from the data producer;

wherein, on an initial iteration of the iterations, the third output is based on the second input; and

wherein, on subsequent iterations of the iterations, the third output is based on the fourth input of an immediately preceding iteration of the iterations; and

transmitting, toward the core microservice, fourth output comprising the fourth input received on a final iteration of the iterations;

the second operations comprising:

receiving a ninth input based on the fourth output;

comparing the ninth input with a graphical model of a knowledge corpus incorporated in the copilot;

determining whether the copilot is competent to act on the ninth input;

upon determining that the copilot is competent:

determining and transmitting ninth output based on the ninth input, toward the core microservice; and

upon determining that the copilot is not competent, transmitting a notification indicating lack of competence;

wherein the ninth output enables the core microservice to generate a response to the first input from the client application.

17. A system comprising:

one or more hardware processors, with memory coupled thereto; and

one or more computer readable media storing instructions comprising a plurality of modules which, when executed by the one or more hardware processors, implement respective microservices, the microservices forming a weakly connected network of microservices configured as a copilot for one or more first client applications;

wherein each of the microservices is configured to:

receive input from (i) a respective first group comprising one or more others of the microservices or (ii) one or more second client applications; and

transmit output to (i) a second group comprising one or more of the microservices or (ii) one or more third client applications;

wherein a plurality of the microservices incorporate respective trained machine learning tools;

wherein the network of microservices comprises at least a retrieval microservice, one or more data producers, and a core microservice;

wherein the retrieval microservice is configured to:

receive second input, which reflects first input from a given one of the one or more first client applications;

recursively perform a plurality of iterations of retrieval augmented generation (RAG), each of the iterations comprising:

transmitting third output to a respective data producer of the one or more data producers, wherein the respective data producer, at a given iteration of the iterations, is a database microservice; and

responsive to the transmitting, receiving fourth input from the respective data producer;

wherein, on an initial iteration of the iterations, the third output is based on the second input; and

wherein, on subsequent iterations of the iterations, the third output is based on the fourth input of an immediately preceding iteration of the iterations; and

transmit, toward the core microservice, fourth output comprising the fourth input received on a final iteration of the iterations;

wherein the fourth output enables the core microservice to generate a response to the first input from the given first client application; and

wherein the database microservice is configured to:

receive, on the given iteration, a seventh input based on the third output;

retrieve, from one or more databases, database objects relevant to the seventh input; and

determine and transmit a seventh output based on the retrieved database objects;

wherein the fourth input corresponding to the third output, on the given iteration, is based on the seventh output.

Assignments (2)

Continuity (3)

References Cited (129)