Granted Patent B1

US 12,657,620 · App. 18/384,819 · Granted Jun 16, 2026

Inventors: Christopher Brossman (San Francisco, CA); Anurag Syal (Fremont, CA)

Assignee: The RealReal, Inc.

G06Q30/0631G06Q10/087

Abstract

Claims (63)

1 . A method comprising:

obtaining event information;

obtaining user information associated with a user;

obtaining available inventory information;

generating one or more items based on the event information, the user information, and the available inventory information; and

presenting the one or more items to the user, wherein generating the one or more items includes using:

an event model configured to generate an event embedding from the event information;

a user model configured to generate a user embedding from the user information;

an inventory model configured to generate available inventory embeddings from the available inventory information; and

a domain matching model configured to generate a domain matching embedding from the event embedding and the user embedding, wherein the domain matching embedding has an equivalent dimension as the available inventory embeddings, and to identify the one or more items by:

analyzing the domain matching embedding and a plurality of the available inventory embeddings to compute cosine similarity values between the domain matching embedding and each available inventory embedding of the plurality of the available inventory embeddings;

in response to detecting the cosine similarity values, ranking the plurality of the available inventory embeddings based on the cosine similarity values; and

in response to detecting a highest-ranked set of the available inventory embeddings, selecting, as the one or more items, a top N set of available inventory items corresponding to highest-ranked available inventory embeddings.

2 . The method of claim 1 , wherein event information includes one or more items selected from the group consisting of: event type information of an event, location information of the event, event setting information, climate information, culture information of a geographic area surrounding the event, and event-related information.

3 . The method of claim 1 , wherein event information is obtained from at least one of large language models (LLMs), generative artificial intelligence (AI) models, stored data, a data scraping system, or user input data.

4 . The method of claim 1 , wherein the generating of the one or more items based on the event information, the user information, and the available inventory information comprises:

using an event model configured to generate an event embedding from the event information;

using a user model configured to generate a user embedding from the user information;

using a domain matching model configured to generate a domain matching embedding from the event embedding and the user embedding;

using an inventory model configured to generate available inventory embeddings from the available inventory information; and

generating the one or more items using the available inventory embeddings and the domain matching embedding.

5 . The method of claim 4 , wherein each of the event embedding, the user embedding, and the available inventory embeddings need not have an equivalent dimension, and wherein the domain matching embedding has an equivalent dimension as the available inventory embeddings.

6 . The method of claim 4 , wherein the generating of the one or more items based on the event information, the user information, and the available inventory information further comprises:

generating a location embedding from the location information;

generating an item embedding from the item information, and wherein the domain matching embedding is generated from the event embedding, the user embedding, the location embedding, and the item embedding.

7 . The method of claim 4 , wherein the generating the one or more items using the available inventory embeddings and the domain matching embedding involves ranking the available inventory embeddings most similar to the domain matching embedding.

8 . The method of claim 1 , wherein the user information indicates buying preferences of the user and includes at least one item selected from the group consisting of: a price range preference, designer or item source preference, taxon preference, color preference, sizing information, gender information, and historical purchasing information.

9 . The method of claim 1 , wherein the available inventory indicates inventory available for purchase from a single SKU (stock keeping unit) provider entity.

10 . A system comprising:

a storage system; and

an event based item generator configured to (i) obtain event information, user information associated with a user, and available inventory information, and (ii) generate one or more items based on the event information, the user information, and the available inventory information, wherein to generate the one or more items, the event based item generator comprises:

an event model configured to generate an event embedding from the event information;

a user model configured to generate a user embedding from the user information;

an inventory model configured to generate available inventory embeddings from the available inventory information; and

a domain matching model configured to generate a domain matching embedding from the event embedding and the user embedding, wherein the domain matching embedding has the same dimension as the available inventory embeddings, and to identify the one or more items by:

analyzing the domain matching embedding and a plurality of the available inventory embeddings to compute cosine similarity values between the domain matching embedding and each available inventory embedding of the plurality of the available inventory embeddings;

in response to detecting the cosine similarity values, ranking the plurality of the available inventory embeddings based on the cosine similarity values; and

in response to detecting a highest-ranked set of the available inventory embeddings, selecting, as the one or more items, a top N set of available inventory items corresponding to highest-ranked available inventory embeddings.

11 . The system of claim 10 , wherein the event based item generator is also configured to (iii) present the one or more items to the user.

12 . The system of claim 10 , wherein event information includes one or more items selected from a group consisting of: event type information of an event, location information of the event, event setting information, climate information, culture information of a geographic area surrounding the event, and event-related information.

13 . The system of claim 10 , wherein the user information indicates buying preferences of the user and includes at least one item selected from the group consisting of: a price range preference, designer or item source preference, taxon preference, color preference, sizing information, gender information, and historical purchasing information.

14 . The system of claim 10 , wherein the one or more items are provided by a single SKU (stock keeping unit) provider.

15 . The system of claim 10 , wherein in generating the one or more items based on the event information, the user information, and the available inventory information, the event based item generator is further configured to:

generating an event embedding from the event information;

generating a user embedding from the user information;

generating a domain matching embedding from the event embedding and the user embedding;

generating available inventory embeddings from the available inventory information; and

generating the one or more items using the available inventory embeddings and the domain matching embedding, wherein the event based item generator comprises an event model configured to generate the event embedding, a user model configured to generate the user embedding, an inventory model configured to generate the available inventory embeddings, and a domain matching model configured to generate the domain matching embedding and to identify the one or more items by:

analyzing the domain matching embedding and a plurality of the available inventory embeddings to compute cosine similarity values between the domain matching embedding and each available inventory embedding of the plurality of the available inventory embeddings;

in response to detecting the cosine similarity values, ranking the plurality of the available inventory embeddings based on the cosine similarity values; and

in response to detecting a highest-ranked set of the available inventory embeddings, selecting, as the one or more items, a top N set of available inventory items corresponding to highest-ranked available inventory embeddings.

16 . The system of claim 15 , wherein in generating the one or more items based on the event information, the user information, and the available inventory information, the event based item generator is further configured to:

generate a location embedding from the location information;

generate an item embedding from the item information, and wherein the domain matching embedding is generated from the event embedding, the user embedding, the location embedding, and the item embedding.

17 . The system of claim 15 , wherein in generating the one or more items based on the event information, the user information, and the available inventory information, the event based item generator ranks the available inventory embeddings most similar to the domain matching embedding.

18 . The system of claim 10 , wherein the event based item generator comprises a computing instance provided on a cloud computing platform.

19 . An event based item generator comprising:

a storage system; and

means for obtaining event information, user information associated with a user, and available inventory information, wherein the means is also for generating one or more items based on the event information, the user information, and the available inventory information, wherein the means identifies the one or more items by:

analyzing a domain matching embedding and a plurality of available inventory embeddings to compute cosine similarity values;

in response to detecting the cosine similarity values, ranking the plurality of available inventory embeddings based on the cosine similarity values; and

in response to detecting a highest-ranked set of available inventory embeddings, selecting a top N set of available inventory items corresponding to highest-ranked available inventory embeddings.

20 . The event based item generator of claim 19 , wherein the means is a computing instance operating on a cloud computing platform.

Assignments (2)

Continuity (1)

References Cited (14)