Learning and personal data management platform

Patent metadata

Field	Value
Publication number	US-12591567-B2
Application number	US-202418809814-A
Country	US
Kind code	B2
Filing date	Aug 20, 2024
Priority date	Aug 23, 2023
Publication date	Mar 31, 2026
Grant date	Mar 31, 2026

Abstract

Methods, systems, and techniques for processing natural language queries. A natural language query related to personal information of the user is obtained. Query vectors that include embeddings of the query are generated and matched with document vectors generated from chunks of reference documents related to the personal data. The document chunks are retrieved from a database and used as context for a prompt to a large language model that is used to respond to the natural language query. The query itself is also included in the prompt. The query may be in respect of a particular goal, and the large language model's response may include recommendations to help the user achieve that goal.

First claim

The invention claimed is:

1 . A natural language query processing method, the method comprising:
(a) obtaining a natural language query related to personal financial data of a user; (b) generating query vectors comprising embeddings of the natural language query; (c) matching the query vectors with document vectors generated from chunks of reference documents related to the personal financial data; (d) retrieving the chunks from at least one database; (e) generating a response to the natural language query by inputting to at least one large language model a prompt comprising:
(i) the natural language query; and
(ii) the chunks as context for use in responding to the natural language query; and
(f) providing the response to the user, wherein the method further comprises, prior to the obtaining of the natural language query, detecting a trigger resulting from a change in the personal financial data of the user, wherein the change comprises at least one of asset or liability values of the user as stored in the at least one database exceeding a trigger threshold, and wherein the obtaining of the natural language query is performed in response to the trigger.

2 . The natural language query processing method of claim 1 , wherein the document vectors are associated in the at least one database with respective identifiers identifying the reference documents used to generate the document vectors, and wherein the method further comprises:
(a) generating a link to at least one of the reference documents, respectively, wherein the generating comprises identifying the at least one of the reference documents corresponding to at least one of the chunks by matching the at least one of the identifiers of the at least one of the reference documents to at least one of the document vectors generated from the at least one of the chunks; and (b) providing the link to the user.

3 . The natural language query processing method of claim 1 , further comprising intermittently updating the at least one of the asset or liability values in the at least one database by collecting the at least one of the asset or liability values through at least one application programming interface (“API”) endpoint providing access to third party services, wherein the collecting comprises accessing the third party services using respective login credentials or biometric identification of the user.

4 . The natural language query processing method of claim 1 , wherein the change comprises the asset values of the user increasing within a single day by more than the trigger threshold.

5 . The natural language query processing method of claim 1 , further comprising, prior to the obtaining of the natural language query:
(a) generating the chunks from the reference documents, wherein the reference documents are associated with respective identifiers; (b) generating the document vectors from the chunks; and (c) storing the document vectors with the identifiers in the at least one database.

6 . The natural language query processing method of claim 1 , further comprising:
(a) obtaining the personal financial data of the user; (b) obtaining at least one goal of the user related to the personal financial data through a graphical user interface; (c) determining that the user has satisfied the at least one goal; and (d) in response to the user satisfying at least one of the at least one goal, crediting the user with points.

7 . The natural language query processing method of claim 6 , wherein each of the at least one goal is stored in the at least one database as a first lookup table, wherein each of the at least one goal comprises at least one task, and wherein each of the at least one task is stored in the at least one database as a second lookup table.

8 . The natural language query processing method of claim 7 , wherein each of the at least one goal comprises part of a user plan, and wherein the user plan is stored in the at least one database as a third lookup table.

9 . The natural language query processing method of claim 6 , wherein the points are allocated by a points provider, and wherein the crediting the user with the points comprises instructing the points provider to credit the user with the points and providing credentials of the user for the points provider to the points provider.

10 . The natural language query processing method of claim 6 , wherein the at least one goal comprises a shared goal that is shared by multiple users of which the user is one, wherein the method further comprises graphically displaying to the user:
(a) a first graphical interface element indicating respective progress towards the shared goal of each of the multiple users; and (b) a second graphical interface element permitting the user to perform a task that causes the user to advance towards the shared goal.

11 . The natural language query processing method of claim 10 , wherein the shared goal is a monetary savings goal, and wherein the task is to contribute a set amount of funds of the user towards the monetary savings goal of the user.

12 . The natural language query processing method of claim 6 , wherein the prompt further comprises the personal financial data of the user and the natural language query is in respect of the at least one goal of the user.

13 . The natural language query processing method of claim 12 , wherein inputting the prompt to the at least one large language model comprises applying a few-shot prompting technique wherein the prompt further comprises an example natural language query and at least one example response, wherein each of the at least one example response comprises a response title and a response body.

14 . The natural language query processing method of claim 13 , wherein the prompt further comprises instructions to format the response in JavaScript Object Notation (JSON) schema, with at least one respective title and body key for at least one response title and at least one response body to be output by the at least one large language model.

15 . The natural language query processing method of claim 14 , wherein the at least one response title and at least one response body provide at least one recommendation to advance towards the at least one goal to the user, and wherein the method further comprises submitting the JSON-formatted response to an API that displays the at least one recommendation to the user.

16 . The natural language query processing method of claim 15 , further comprising displaying to the user on the same screen as the at least one recommendation, a graphical interface element permitting the user to perform a task that causes the user to advance towards the at least one goal.

17 . The natural language query processing method of claim 6 , further comprising:
(a) displaying to the user an interactive learning module related to the at least one goal; and (b) following completion of the interactive learning module by the user, crediting the user with additional ones of the points, wherein the points are allocated by a points provider, and wherein the crediting the user with the points comprises instructing the points provider to credit the user with the points and providing credentials of the user for the points provider to the points provider.

18 . A system comprising:
(a) at least one network interface; (b) at least one processing unit communicatively coupled to the at least one network interface and configured to perform a natural language query processing method, the method comprising:
(i) obtaining a natur

Assignees

• Royal Bank Of Canada

Inventors

• Currell Nathan
• Luu Hilton
• Zhang Linna
• Ghassel Abdellah

Classifications

• G06F16/243 Primary

  Natural language query formulation · CPC title