IP Library › Granted Patent US 12,699,836

Granted Patent B2

US 12,699,836 · App. 18/646,500 · Granted Aug 4, 2026

Content generation using enhanced actor-critic models

Inventors: Yifan Xu (Redmond, WA); Galina Alperovich (Prague, CZ); Varsha Mahadevan (Bengaluru, IN)

G06F40/186 G06F40/40

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 12,699,836
App. No.: 18/646,500
Granted: Aug 4, 2026
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

Methods, systems, and devices for content generation are described. Techniques described herein may support the use of multiple large language model (LLM) agents to generate content. For example, a first actor agent may generate a layout including one or more sections to fulfill a user-supplied query. A second actor agent may retrieve information, and a first critic agent may review the retrieved information to verify that the retrieved information fulfills the generated layout. If the retrieved information does not fulfill the layout, the second actor agent may obtain additional information. A third actor agent may use the retrieved information to generate a draft query response. A second critic agent may review the draft query response to determine if the draft query response is in accordance with user-supplied guidelines. The second actor agent may revise the query response if the query response is not in accordance with the user-supplied guidelines.

Claims (63)

1 . A method for content generation, comprising:

obtaining, at a content generation system comprising a plurality of agents, a user input comprising a plain-text query;

generating, by a first actor agent of the plurality of agents and using the plain-text query, a layout comprising one or more sections to be augmented using content obtained from one or more information sources;

obtaining, by a second actor agent after generating the plain-text query and using the layout and the plain-text query, the content comprising information from one or more information sources;

outputting, by a first critic agent using the layout and the content as input, a message indicating that the content supports the one or more sections of the layout;

generating, by a third actor agent after the message is output by the first critic agent, a draft query response using at least a portion of the content and the layout;

outputting, by a second critic agent and using one or more input prompts comprising content guidelines, one or more plain-text messages related to a quality of the draft query response; and

outputting, by the third actor agent and using the one or more plain-text messages, a revised draft query response.

2 . The method of claim 1 , further comprising:

outputting, by the first critic agent, a message indicating that the content does not support at least one of the one or more sections of the layout; and

obtaining, by the second actor agent, additional content from the one or more information sources in accordance with the message.

3 . The method of claim 1 , further comprising:

generating, by the third actor agent, the revised draft query response in accordance with the one or more plain-text messages; and

outputting, by the second critic agent, one or more additional plain-text messages related to a quality of the revised draft query response in accordance with the one or more input prompts comprising the content guidelines.

4 . The method of claim 1 , further comprising:

performing, by the second actor agent, a check procedure of the content, wherein performing the check procedure comprises determining whether the information is included in at least two of the one or more information sources.

5 . The method of claim 1 , further comprising:

storing, by the second actor agent, the content in a first memory, wherein the first memory is accessible by the second actor agent, the first critic agent, and the third actor agent; and

processing, by the first critic agent after the content is stored in the first memory, the content to determine whether the content supports the one or more sections of the layout.

6 . The method of claim 1 , further comprising:

storing, by the third actor agent, the draft query response in a second memory, wherein the second memory is accessible by the third actor agent and the second critic agent; and

processing, by the second critic agent after the content is stored in the second memory, the draft query response for generating the one or more plain-text messages.

7 . The method of claim 1 , wherein generating the draft query response comprises:

generating, by the third actor agent, the draft query response using both the one or more input prompts comprising the content guidelines and the at least the portion of the content.

8 . The method of claim 1 , wherein the one or more input prompts comprising the content guidelines comprise grammar guidelines, language to be included from the draft query response, language to be excluded from the draft query response, formatting guidelines, or any combination thereof.

9 . The method of claim 1 , wherein the one or more information sources comprise one or more publicly available information sources, one or more internal information sources, or both.

10 . The method of claim 1 , wherein the first actor agent, the second actor agent, the third actor agent, the first critic agent, the second critic agent, or any combination thereof are respective derivations of a large language model.

11 . An apparatus for content generation, comprising:

one or more memories storing processor-executable code; and

one or more processors coupled with the one or more memories and individually or collectively operable to execute the code to cause the apparatus to:

obtain, at a content generation system comprising a plurality of agents, a user input comprising a plain-text query;

generate, by a first actor agent of the plurality of agents and using the plain-text query, a layout comprising one or more sections to be augmented using content obtained from one or more information sources;

obtain, by a second actor agent after generating the plain-text query and using the layout and the plain-text query, the content comprising information from one or more information sources;

output, by a first critic agent using the layout and the content as input, a message indicating that the content supports the one or more sections of the layout;

generate, by a third actor agent after the message is output by the first critic agent, a draft query response using at least a portion of the content and the layout;

output, by a second critic agent and using one or more input prompts comprising content guidelines, one or more plain-text messages related to a quality of the draft query response; and

output, by the third actor agent and using the one or more plain-text messages, a revised draft query response.

12 . The apparatus of claim 11 , wherein the one or more processors are individually or collectively further operable to execute the code to cause the apparatus to:

output, by the first critic agent, a message indicating that the content does not support at least one of the one or more sections of the layout; and

obtain, by the second actor agent, additional content from the one or more information sources in accordance with the message.

13 . The apparatus of claim 11 , wherein the one or more processors are individually or collectively further operable to execute the code to cause the apparatus to:

generate, by the third actor agent, the revised draft query response in accordance with the one or more plain-text messages; and

output, by the second critic agent, one or more additional plain-text messages related to a quality of the revised draft query response in accordance with the one or more input prompts comprising the content guidelines.

14 . The apparatus of claim 11 , wherein the one or more processors are individually or collectively further operable to execute the code to cause the apparatus to:

perform, by the second actor agent, a check procedure of the content, wherein performing the check procedure comprises determining whether the information is included in at least two of the one or more information sources.

15 . The apparatus of claim 11 , wherein the one or more processors are individually or collectively further operable to execute the code to cause the apparatus to:

store, by the second actor agent, the content in a first memory, wherein the first memory is accessible by the second actor agent, the first critic agent, and the third actor agent; and

process, by the first critic agent after the content is stored in the first memory, the content to determine whether the content supports the one or more sections of the layout.

16 . The apparatus of claim 11 , wherein the one or more processors are individually or collectively further operable to execute the code to cause the apparatus to:

store, by the third actor agent, the draft query response in a second memory, wherein the second memory is accessible by the third actor agent and the second critic agent; and

process, by the second critic agent after the content is stored in the second memory, the draft query response for generating the one or more plain-text messages.

17 . The apparatus of claim 11 , wherein, to generate the draft query response, the one or more processors are individually or collectively operable to execute the code to cause the apparatus to:

generate, by the third actor agent, the draft query response using both the one or more input prompts comprising the content guidelines and the at least the portion of the content.

18 . The apparatus of claim 11 , wherein the one or more input prompts comprising the content guidelines comprise grammar guidelines, language to be included from the draft query response, language to be excluded from the draft query response, formatting guidelines, or any combination thereof.

19 . The apparatus of claim 11 , wherein the one or more information sources comprise one or more publicly available information sources, one or more internal information sources, or both.