Granted Patent B2

US 12,651,025 · App. 19/175,928 · Granted Jun 9, 2026

Inventors: Michele Riva (San Francisco, CA); Tommaso Allevi (Milan, IT); Issac Roth (San Francisco, CA)

Assignee: Oramasearch Inc.

G06F16/9027G06F16/93

Abstract

Claims (86)

1 . One or more memory devices collectively having contents configured to cause a computing system to perform a method, the method comprising:

accessing one or more training documents, each of the training documents having contents;

accessing, for each of the training documents, a list of properties of the training document that should be indexed;

for each of the accessed training documents, constructing a training example comprising (1) the training document's contents and (2) the list of properties of the training document that should be indexed;

accessing a corpus of documents to be indexed;

constructing and submitting a textual prompt to a large language model, wherein the textual prompt specifies (1) the constructed training examples, (2) the accessed corpus of documents, (3) an indication that the training examples should be used as a basis to identify in the corpus of documents properties that should be used to index documents of the corpus, and (4) a request for a response that specifies at least the identified properties that should be used to index documents of the corpus;

receiving from the large language model the response to the prompt that identifies, for each of the documents of the corpus, a list of properties that are automatically selected by the large language model and proposed to be a basis for an index should be included in an index on the corpus for the document of the corpus; and

selecting the properties identified in the response received from the large language model to construct, at least in part, an index on the documents of the corpus.

2 . The one or more memory devices of claim 1 wherein the accessed lists of properties include properties that would be common to search on or filter on for the document of the corpus.

3 . The one or more memory devices of claim 1 , the method further comprising:

constructing the index on the documents of the corpus that is based on the selected properties.

4 . The one or more memory devices of claim 3 , the method further comprising:

before the constructing:

causing the properties identified in the response received from the large language model to be presented to a user; and

receiving input originated by the user:

selecting one or more unselected document properties,

unselecting one or more selected document properties, or

selecting one or more unselected document properties and unselecting one or more selected document properties; and

adjusting selection among the document properties in accordance with the received input.

5 . The one or more memory devices of claim 3 wherein the constructed index is a tree structure mapping from the selected properties to the documents of the corpus having those properties.

6 . The one or more memory devices of claim 3 wherein the constructing comprises:

for each document of the corpus,

concatenating the document's contents of the selected properties into a representation of the document;

encoding the representation of the document into a vector wherein the constructed index maps from the vectors produced by the encoding to the corresponding document of the corpus.

7 . The one or more memory devices of claim 3 , the method further comprising:

using the constructed index to conduct a search of the corpus.

8 . The one or more memory devices of claim 3 , the method further comprising:

using the constructed index to service a chatbot interaction with respect to the corpus.

9 . A computer-implemented method comprising:

accessing one or more training documents, each of the training documents having contents;

accessing, for each of the training documents, a list of properties of the training document that should be indexed;

for each of the accessed training documents, constructing a training example comprising (1) the training document's contents and (2) the list of properties of the training document that should be indexed;

accessing a corpus of documents to be indexed;

constructing and submitting a textual prompt to a large language model, wherein the textual prompt specifies (1) the constructed training examples, (2) the accessed corpus of documents, (3) an indication that the training examples should be used as a basis to identify in the corpus of documents properties that should be used to index documents of the corpus, and (4) a request for a response that specifies at least the identified properties that should be used to index documents of the corpus;

receiving from the large language model the response to the prompt that identifies, for each of the documents of the corpus, a list of properties that are automatically selected by the large language model and proposed to be a basis for an index should be included in an index on the corpus for the document of the corpus; and

selecting the properties identified in the response received from the large language model to construct, at least in part, an index on the documents of the corpus.

10 . The method of claim 9 wherein the accessed lists of properties include properties that would be common to search on or filter on for the document of the corpus.

11 . The method of claim 9 , further comprising:

constructing the index on the documents of the corpus that is based on the selected properties.

12 . The method of claim 11 , further comprising:

before the constructing:

causing the properties identified in the response received from the large language model to be presented to a user; and

receiving input originated by the user:

selecting one or more unselected document properties,

unselecting one or more selected document properties, or

selecting one or more unselected document properties and unselecting one or more selected document properties; and

adjusting selection among the document properties in accordance with the received input.

13 . The method of claim 11 wherein the constructed index is a tree structure mapping from the selected properties to the documents of the corpus having those properties.

14 . The method of claim 11 wherein the constructing comprises:

for each document of the corpus,

concatenating the document's contents of the selected properties into a representation of the document;

encoding the representation of the document into a vector wherein the constructed index maps from the vectors produced by the encoding to the corresponding document of the corpus.

15 . The method of claim 11 , further comprising:

using the constructed index to conduct a search of the corpus.

16 . The method of claim 11 , further comprising:

using the constructed index to service a chatbot interaction with respect to the corpus.

17 . A system, comprising:

one or more processors; and

memory storing content that, when executed by the one or more processors, cause the system to perform a method comprising:

accessing one or more training documents, each of the training documents having contents;

accessing, for each of the training documents, a list of properties of the training document that should be indexed;

for each of the accessed training documents, constructing a training example comprising (1) the training document's contents and (2) the list of properties of the training document that should be indexed;

accessing a corpus of documents to be indexed;

constructing and submitting a textual prompt to a large language model, wherein the textual prompt specifies (1) the constructed training examples, (2) the accessed corpus of documents, (3) an indication that the training examples should be used as a basis to identify in the corpus of documents properties that should be used to index documents of the corpus, and (4) a request for a response that specifies at least the identified properties that should be used to index documents of the corpus;

receiving from the large language model the response to the prompt that identifies, for each of the documents of the corpus, a list of properties that are automatically selected by the large language model and proposed to be a basis for an index should be included in an index on the corpus for the document of the corpus; and

selecting the properties identified in the response received from the large language model to construct, at least in part, an index on the documents of the corpus.

18 . The system of claim 17 wherein the accessed lists of properties include properties that would be common to search on or filter on for the document of the corpus.

19 . The system of claim 17 , the method further comprising:

constructing the index on the documents of the corpus that is based on the selected properties.

20 . The system of claim 19 , the method further comprising:

before the constructing:

causing the properties identified in the response received from the large language model to be presented to a user; and

receiving input originated by the user:

selecting one or more unselected document properties,

unselecting one or more selected document properties, or

selecting one or more unselected document properties and unselecting one or more selected document properties; and

adjusting selection among the document properties in accordance with the received input.

21 . The system of claim 19 wherein the constructed index is a tree structure mapping from the selected properties to the documents of the corpus having those properties.

22 . The system of claim 19 wherein the constructing comprises:

for each document of the corpus,

concatenating the document's contents of the selected properties into a representation of the document;

encoding the representation of the document into a vector wherein the constructed index maps from the vectors produced by the encoding to the corresponding document of the corpus.

23 . The system of claim 19 , the method further comprising:

using the constructed index to conduct a search of the corpus.

24 . The system of claim 19 , the method further comprising:

using the constructed index to service a chatbot interaction with respect to the corpus.

Assignments (2)

Continuity (2)

References Cited (29)