IP Library › Granted Patent US 12,699,850

Granted Patent B2

US 12,699,850 · App. 18/759,536 · Granted Aug 4, 2026

Automated response system with API calls and human agent interaction via language model prompts

Inventors: Tao Ma (Mountain View, CA); Hugh Nicholas Perkins (New York, NY); Mitchell Steven Morgan (Cedar Park, TX); Michael Griffiths (Brooklyn, NY); Satchuthananthavale Rasiah Kuhan Branavan (Caterham, GB); Kilian Quirin Weinberger (Ithaca, NY)

Assignee: ASAPP, INC.

G06F40/35 G06F9/543 G06F40/40 H04M3/5191

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 12,699,850
App. No.: 18/759,536
Granted: Aug 4, 2026
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

An automated support process implemented with a language model may be improved by allowing a human agent to provide input to the language model to fix mistakes or otherwise improve the automated support process. In some implementations, a language model prompt may be used to instruct a language model to be able to provide a response that includes requesting assistance from a human agent. The request for assistance may then be sent to the human agent and a response from the human agent may then be processed by the language model to allow the language model to provide better responses for the support process. In some implementations, a human agent may monitor automated support sessions and provide unrequested input into a support session where needed to improve the quality of the support session.

Claims (90)

1 . A computer-implemented method, comprising:

receiving a first user communication from a first user;

generating a first language model prompt, wherein the first language model prompt:

comprises a representation of the first user communication,

describes a plurality of language model response types, the plurality of language model response types comprising (a) transmitting a communication to a user, (b) performing an API call, and (c) requesting assistance from a human agent, and

instructs a language model to return a language model response;

receiving a first language model response from the language model as a result of submitting the first language model prompt to the language model, wherein the first language model response indicates a type of performing an API call and comprises a representation of a first API call;

receiving a first API response as a result of performing the first API call;

generating a second language model prompt, wherein the second language model prompt:

comprises the representation of the first user communication and a representation of the first API response,

describes the plurality of language model response types, and

instructs the language model to return a language model response;

receiving a second language model response from the language model as a result of submitting the second language model prompt to the language model, wherein the second language model response indicates a type of requesting assistance from a human agent and comprises a representation of a first human agent request;

causing the first human agent request to be presented to a first human agent;

receiving a first human agent communication from the first human agent;

generating a third language model prompt, wherein the third language model prompt:

comprises the representation of the first user communication, the representation of the first API response, and a representation of the first human agent communication,

describes the plurality of language model response types, and

instructs the language model to return a language model response;

receiving a third language model response from the language model as a result of submitting the third language model prompt to the language model, wherein the third language model response indicates a type of transmitting a communication to a user and comprises a representation of a first automated communication; and

causing the first automated communication to be presented to the first user.

2 . The computer-implemented method of claim 1 , wherein the third language model prompt comprises the representation of the first API call and the representation of the first human agent request.

3 . The computer-implemented method of claim 1 , wherein the first language model prompt describes a plurality of API calls.

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

maintaining a first context for using the language model to communicate with the first user, wherein the first context comprises the representation of the first user communication; and

maintaining a second context for using the language model to communicate with the first human agent, wherein the second context comprises the first human agent communication.

5 . The computer-implemented method of claim 4 , comprising maintaining a third context for using the language model to communicate with an API gateway.

6 . The computer-implemented method of claim 4 , comprising copying a portion of the second context to the first context.

7 . The computer-implemented method of claim 1 , wherein the first user communication is a text communication.

8 . A system, comprising at least one server computer comprising at least one processor and at least one memory, the at least one server computer configured to:

receive a first user communication from a first user;

generate a first language model prompt, wherein the first language model prompt:

comprises a representation of the first user communication,

instructs a language model to return a language model response;

receive a first language model response from the language model as a result of submitting the first language model prompt to the language model, wherein the first language model response indicates a type of performing an API call and comprises a representation of a first API call;

receive a first API response as a result of performing the first API call;

generate a second language model prompt, wherein the second language model prompt:

comprises the representation of the first user communication and a representation of the first API response,

describes the plurality of language model response types, and

instructs the language model to return a language model response;

receive a second language model response from the language model as a result of submitting the second language model prompt to the language model, wherein the second language model response indicates a type of requesting assistance from a human agent and comprises a representation of a first human agent request;

cause the first human agent request to be presented to a first human agent;

receive a first human agent communication from the first human agent;

generate a third language model prompt, wherein the third language model prompt:

comprises the representation of the first user communication, the representation of the first API response, and a representation of the first human agent communication,

describes the plurality of language model response types, and

instructs the language model to return a language model response;

receive a third language model response from the language model as a result of submitting the third language model prompt to the language model, wherein the third language model response indicates a type of transmitting a communication to a user and comprises a representation of a first automated communication; and

cause the first automated communication to be presented to the first user.

9 . The system of claim 8 , wherein the at least one server computer is configured to:

maintain a first context for using the language model to communicate with the first user, wherein the first context comprises the representation of the first user communication; and

maintain a second context for using the language model to communicate with the first human agent, wherein the second context comprises the first human agent communication.

10 . The system of claim 9 , wherein the at least one server computer is configured to copy a portion of the second context to the first context.

11 . The system of claim 8 , wherein the first user communication is an audio communication and the representation of the first user communication corresponds to a transcription of the first user communication.

12 . The system of claim 8 , wherein the at least one server computer is configured to:

receive a second human agent communication from the first human agent, wherein the second human agent communication is not in response to a human agent request; and

generate a fourth language model prompt, wherein the fourth language model prompt comprises a representation of the second human agent communication.

13 . The system of claim 8 , wherein the at least one server computer is configured to:

determine a task category of the first user communication; and

generate the first language model prompt using the task category of the first user communication.

14 . The system of claim 13 , wherein the at least one server computer is configured to determine the task category of the first user communication by processing the first user communication with an intent classifier.

15 . The system of claim 13 , wherein the at least one server computer is configured to determine the task category of the first user communication by generating a language model prompt and receiving a language model response from the language model or another language model.

16 . One or more non-transitory, computer-readable media comprising computer-executable instructions that, when executed, cause at least one processor to perform actions comprising: