US 12,608,417 B1Granted

Artificial intelligence language model orchestrator for interactive searching

AssigneeAmazon Technologies, Inc.

InventorsArnab Dhua, Oriol Barbany Mayor, Sheng-Wei Huang, Xinliang Zhu, Anuj Bindal

Life of this patent

Estimating the term…

Loading patent…

Abstract

An artificial intelligence (“AI”) search system includes a generative model orchestrator. A user provides an initial input query. The AI search system can determine initial search results. The AI search system receives subsequent input queries, which can include user input indicative of user intent in natural language, such as, altering, complementary, reinforcing, and unrelated intent. The AI search system creates input data that includes a prefix, format instructions, example(s), the input query, and/or contextual metadata. The AI search system provides the input data to the generative model orchestrator, which can, as instructed, output an action for a service, such as a search or image service. The AI search system executes the action, determines additional search results, and provide the additional search result(s).

Claims20

Dependent claims

A system comprising:

a non-transitory data storage medium to store specific computer-executable instructions;
and
a computer hardware processor in communication with the non-transitory data storage medium, wherein the computer hardware processor is to execute the specific computer-executable instructions to at least:
receive, via a user interface, a first query from a user computing device, the first query comprising an input image;
determine a first search result associated with a first item based at least in part on the input image;
provide the first search result to the user computing device;
receive, via the user interface, a second query from a user computing device, the second query comprising first input data indicating an attribute change with respect to the first item;
determine, from a database, item metadata from at least one of the first query, the first search result, or the first item;
create second input data based at least in part on (i) a set of instructions to output an action, (ii) the first input data, and (iii) the item metadata;
provide the second input data to a generative model orchestrator, wherein the generative model orchestrator outputs an output action for a first service from a plurality of services, wherein the first service comprises a search service, and the output action indicates the attribute change with respect to the input image;
invoke the output action for the first service, wherein the first service returns a second search result;
and
provide the second search result to the user computing device.

10 more elements

A computer-implemented method comprising:

receiving, via a user interface, a first query from a user computing device;
determine a first search result associated with a first item based at least in part on the first query;
providing the first search result to the user computing device;
receiving, via the user interface, a second query from a user computing device, the second query comprising first input data indicating user intent for a search relative to the first item;
determining contextual metadata from at least one of the first query, the first search result, or the first item;
creating second input data based at least in part on (i) a set of instructions to output an action, (ii) the first input data, and (iii) the contextual metadata;
providing the second input data to a generative model orchestrator, wherein the generative model orchestrator outputs an output action for a first service from a plurality of services;
invoking the output action for the first service, wherein the first service returns service output;
determining a second search result based at least in part on the service output;
and
providing the second search result to the user computing device.

8 more elements

A system comprising:

a non-transitory data storage medium to store specific computer-executable instructions;
and
a computer hardware processor in communication with the non-transitory data storage medium, wherein the computer hardware processor is to execute the specific computer-executable instructions to at least:
receive, via a user interface, a first query from a user computing device;
determine a first search result associated with a first item based at least in part on the first query;
provide the first search result to the user computing device;
receive, via the user interface, a second query from a user computing device, the second query comprising first input data indicating user intent for a search relative to the first item;
determine, from a database, contextual metadata from at least one of the first query, the first search result, or the first item;
create second input data based at least in part on (i) a set of instructions to output an action, (ii) the first input data, and (iii) the contextual metadata;
provide the second input data to a generative model orchestrator, wherein the generative model orchestrator outputs an output action for a first service from a plurality of services;
invoke the output action for the first service, where the first service returns service output;
determine a second search result based at least in part on the service output;
and
provide the second search result to the user computing device.

11 more elements

Dependencies are inferred from the claim text.