Generative artificial intelligence (AI) driven similarity search

1 . A method for performing a similarity search assisted by a generative artificial intelligence (AI) model, the method comprising operations executed using at least one processor of a computing device, and the operations comprising:

receiving a request for a similarity search, wherein the request for the similarity search is provided in response to a user interaction received in a user interface, and wherein the request for the similarity search includes a product identifier of a selected product already displayed in the user interface;

generating a search query to perform the similarity search, using at least one AI model, wherein the at least one AI model generates the search query based on multiple attributes associated with the selected product, wherein the multiple attributes are identified by the at least one AI model with operations including:

providing the product identifier as input to a first application programming interface (API);

invoking the at least one AI model via the first API, to analyze product data associated with the product identifier and generate multiple attributes of the selected product; and

receiving a generated search query, via the first APL produced from the multiple attributes:

performing the similarity search with the generated search query on a product catalog, without a user provided command to invoke the generated search query with operations including;

providing the generated search query as input to a second API;

invoking the second API to automatically perform searches on the product catalog using the generated search query; and

receiving similarity search results, via the second API based on use of the generated search query;

sorting the similarity search results in response to performing the similarity search on the product catalog; and

presenting the similarity search results in the user interface, wherein the similarity search results include one or more products identified as similar to the selected product.

2 . The method of claim 1 , wherein the user interaction is a request from a user for a new search to present new products more like the selected product, wherein the product identifier is a stock keeping unit (SKU) number, and wherein the user interaction is received with a button or link presented in the user interface to initiate the similarity search based on the selected product.

3 . The method of claim 1 , the operations further comprising:

applying one or more constraints to limit results in the similarity search results, wherein the one or more constraints define requirements for selection of:

one or more products of a same product type as the selected product;

one or more products of a same product brand as the selected product;

one or more products within a predetermined price range above or below the selected product; or

one or more products with an in-stock availability.

4 . The method of claim 3 , the operations further comprising:

evaluating a number of the results in the similarity search results, in response to applying the one or more constraints; and

relaxing an application of the one or more constraints to increase the number of the results in the similarity search, in response to the number of the results not satisfying a minimum number of results.

5 . The method of claim 1 , the operations further comprising, before receiving the request for the similarity search:

receiving, in the user interface, a search request of the product catalog, the search request including at least one keyword or phrase received in the user interface;

generating a product search query to perform the search request, using the at least one AI model; and

presenting, in the user interface, a plurality of product search results in response to the product search query;

wherein the plurality of product search results includes the selected product, and wherein the request for the similarity search is performed with a user selection received in the plurality of product search results.

6 . The method of claim 5 , wherein the at least one AI model identifies the multiple attributes to perform the similarity search based on additional information including (i) the at least one keyword or phrase associated with the search request, (ii) a search intent associated with the search request, and (iii) user activity associated with the search request.

7 . The method of claim 1 , wherein presenting the similarity search results in the user interface includes:

presenting, in a product listing page, a listing of the one or more products identified as similar to the selected product; and

presenting, in a top section of the product listing page, an identification of the selected product as an anchor product.

8 . The method of claim 1 , wherein the at least one AI model includes a generative large language model, and wherein the at least one AI model evaluates the multiple attributes associated with the selected product to determine a search intent, and generates at least one keyword or phrase used for identifying the similarity search results based on the determined search intent.

9 . The method of claim 1 , wherein the at least one AI model includes a predictive model, and wherein the predictive model generates or refines the search query to perform the similarity search based on dynamic context information associated with:

profile information associated with a user;

interactions observed in the user interface; or

search intent associated with an earlier search performed in the user interface.

10 . The method of claim 1 , the operations further comprising, after presenting the similarity search results in the user interface:

presenting one or more search refinement options applicable to the similarity search results;

receiving a selection of the one or more search refinement options;

generating narrowed search results based on applying the selected refinement options to the similarity search results, and

outputting the narrowed search results to be presented in the user interface.

11 . A non-transitory computer-readable storage medium, the computer-readable storage medium comprising instructions that, when executed by processor circuitry and memory of a computing device, causes the computing device to provide similarity search results via a user interface, with operations that:

receive a request for a similarity search, wherein the request for the similarity search is provided in response to a user interaction received in a user interface, and wherein the request for the similarity search includes a product identifier of a selected product already displayed in the user interface;

generate a search query to perform the similarity search, using at least one AI model, wherein the at least one AI model generates the search query based on multiple attributes associated with the selected product, wherein the multiple attributes are identified by the at least one AI model with operations that:

provide the product identifier as input to a first application programming interface (API);

invoke the at least one AI model via the first API, to analyze product data associated with the product identifier and generate multiple attributes of the selected product; and

receive a generated search query, via the first API, produced from the multiple attributes;

perform the similarity search with the generated search query on a product catalog, without a user-provided command to invoke the generated search query, with operations that:

provide the generated search query as input to a second API;

invoke the second API to automatically perform searches on the product catalog using the generated search query; and

receive similarity search results, via the second API, based on use of the generated search query;

sort the similarity search results in response to performing the similarity search on the product catalog; and

present the similarity search results in the user interface, wherein the similarity search results include one or more products identified as similar to the selected product.

12 . The non-transitory computer-readable storage medium of claim 11 , wherein the user interaction is a request from a user for a new search to present new products more like the selected product, wherein the product identifier is a stock keeping unit (SKU) number, and wherein the user interaction is received with a button or link presented in the user interface to initiate the similarity search based on the selected product.

13 . The non-transitory computer-readable storage medium of claim 11 , wherein the instructions further cause the computing device to perform operations that:

apply one or more constraints to limit results in the similarity search results, wherein the one or more constraints define requirements for selection of:

one or more products of a same product type as the selected product;

one or more products of a same product brand as the selected product;

one or more products within a predetermined price range above or below the selected product; or

one or more products with an in-stock availability.

14 . The non-transitory computer-readable storage medium of claim 13 , wherein the instructions further cause the computing device to perform operations that:

evaluate a number of the results in the similarity search results, in response to applying the one or more constraints; and

relax an application of the one or more constraints to increase the number of the results in the similarity search, in response to the number of the results not satisfying a minimum number of results.

15 . The non-transitory computer-readable storage medium of claim 11 , wherein the instructions further cause the computing device to perform operations that, before receipt of the request for the similarity search:

receive, in the user interface, a search request of the product catalog, the search request including at least one keyword or phrase received in the user interface;

generate a product search query to perform the search request, using the at least one AI model; and

present, in the user interface, a plurality of product search results in response to the product search query;

wherein the plurality of product search results includes the selected product, and wherein the request for the similarity search is performed with a user selection received in the plurality of product search results.

16 . The non-transitory computer-readable storage medium of claim 15 , wherein the at least one AI model generates the multiple attributes to perform the similarity search based on additional information including (i) the at least one keyword or phrase associated with the search request, (ii) a search intent associated with the search request, and (iii) user activity associated with the search request.

17 . The non-transitory computer-readable storage medium of claim 11 , wherein to present the similarity search results in the user interface includes to:

present, in a product listing page, a listing of the one or more products identified as similar to the selected product; and

present, in a top section of the product listing page, an identification of the selected product as an anchor product.

18 . The non-transitory computer-readable storage medium of claim 11 , wherein the at least one AI model includes a generative large language model, and wherein the at least one AI model evaluates the multiple attributes associated with the selected product to determine a search intent, and generates at least one keyword or phrase used for identifying the similarity search results based on the determined search intent.

19 . The non-transitory computer-readable storage medium of claim 11 , wherein the at least one AI model includes a predictive model, and wherein the predictive model generates or refines the search query to perform the similarity search based on dynamic context information associated with:

profile information associated with a user;

interactions observed in the user interface; or

search intent associated with an earlier search performed in the user interface.

20 . The non-transitory computer-readable storage medium of claim 11 , wherein the instructions further cause the computing device to perform operations that, after presenting the similarity search results in the user interface:

present one or more search refinement options applicable to the similarity search results;

receive a selection of the one or more search refinement options;

generate narrowed search results based on applying the selected refinement options to the similarity search results; and

output the narrowed search results to be presented in the user interface.