Using generative models for analytic tasks

Patent metadata

Field	Value
Publication number	US-2025348478-A1
Application number	US-202519204433-A
Country	US
Kind code	A1
Filing date	May 9, 2025
Priority date	May 12, 2024
Publication date	Nov 13, 2025
Grant date	—

Abstract

Implementations are provided for facilitating multi-turn dialogs with a generative model-based agent (GMAgent) that allow for multi-step analysis of external data source(s), including refinement of that analysis. In various implementations, data indicative of a first query and external data source(s) may be assembled into a first prompt. The first prompt may be processed using generative model(s) to generate first output data that includes first source code that is executable to perform an analytic task on data from the external data source(s). The first source code may be executed to perform the analytic task using the external data source(s) and generate analytic output. The analytic output may be assembled into a second prompt with a command to determine whether the analytic output satisfies the first query. The second prompt may be processed using generative model(s) to generate second output data that indicates whether the analytic output satisfies the first query.

First claim

What is claimed is:

1 . A method implemented using one or more processors and comprising:
assembling data indicative of a first query and data identifying one or more external data sources into a first prompt; processing the first prompt using one or more generative models to generate first output data, wherein the first output data includes first source code that is executable to extract at least some data from one or more of the external data sources; causing the first source code to be executed to extract at least some data from one or more of the external data sources; assembling the extracted data into a second prompt; processing the second prompt using one or more of the generative models to generate second output data that includes second source code, wherein the second source code is executable to perform an analytic task on data from one or more of the external data sources; and causing the second source code to be executed to perform the analytic task using one or more of the external data sources and generate analytic output.

2 . The method of claim 1 , further comprising causing the analytic output to be presented as output at one or more output devices.

3 . The method of claim 1 , further comprising causing solicitation to be presented at one or more of the output devices, wherein the solicitation presents one or more suggested candidate analytic tasks to be performed on data from one or more of the external data sources, wherein the one or more suggested candidate analytic tasks includes the analytic task that is performed by causing the second source code to be executed.

4 . The method of claim 1 , wherein causing the first source code to be executed results in a subset of data being extracted from one or more of the external data sources.

5 . The method of claim 4 , wherein one or more of the external data sources comprises tabular data, and causing the first source code to be executed results in data from a selected number of rows and/or columns of the tabular data being extracted.

6 . The method of claim 5 , wherein the number of rows and/or columns is selected based on a total number of rows and/or columns in the tabular data.

7 . The method of claim 1 , wherein the analytic output is operable to render an interactive visualization associated with performance of the analytic task.

8 . The method of claim 7 , wherein the analytic output comprises markup language that is operable to render a chart or graphic visualizing one or more aspects of the analytic output or analytic task.

9 . The method of claim 8 , further comprising:
receiving an indication of user interaction with a portion of the chart or graphic; and altering a portion of the markup language that corresponds to the interacted-with portion of the chart or graphic.

10 . The method of claim 9 , further comprising processing a second query, wherein the portion of the markup language is altered based on the second query.

11 . The method of claim 10 , further comprising:
generating a second prompt that includes data indicative of the second query and the portion of the markup language that corresponds to the interacted-with portion of the chart or graphic; and processing the second prompt using one or more of the generative models to generate fifth output indicative of updated markup language that is operable to render an updated chart or graphic.

12 . The method of claim 1 , wherein one or more of the external data sources comprises an uploaded file.

13 . The method of claim 1 , wherein the second source code is executable to perform the analytic task on in-place data within one or more of the external data sources.

14 . The method of claim 1 , further comprising, prior to assembling the first prompt:
assembling data identifying the one or more external data sources into a preliminary prompt; processing the preliminary prompt using one or more of the generative models to generate preliminary output, wherein the preliminary output includes a command to generate the first source code.

15 . The method of claim 1 , further comprising:
assembling the analytic output into a third prompt with a command to determine whether the analytic output satisfies the first query; and processing the third prompt using one or more of the generative models to generate third output data that indicates the analytic output fails to satisfy the first query.

16 . The method of claim 15 , wherein the third output data comprises a command to generate third source code that comprises an updated version of the second source code.

17 . The method of claim 15 , further comprising:
assembling the third output data into a fourth prompt; processing the fourth prompt using one or more of the generative models to generate fourth output data that includes the third source code, wherein the third source code comprises an updated version of the second source code that is executable to perform the analytic task on data from one or more of the external data sources.

18 . A method implemented using one or more processors and comprising:
assembling data identifying one or more external data sources into a first prompt; processing the first prompt using one or more generative models to generate first output data, wherein the first output data includes a command to generate first source code that is executable to extract at least some data from one or more of the external data sources; assembling the command and the data identifying one or more external data sources into a second prompt; processing the second prompt using one or more generative models to generate second output data, wherein the second output data includes the first source code that is executable to extract at least some data from one or more of the external data sources; causing the first source code to be executed to extract data from one or more of the external data sources; assembling the extracted data into a third prompt; processing the third prompt using one or more of the generative models to generate third output data that includes second source code, wherein the second source code is executable to perform an analytic task on data from one or more of the external data sources; causing the second source code to be executed to perform the analytic task using one or more of the external data sources to generate analytic output; and causing the analytic output to be presented as output at one or more output devices.

19 . A method implemented using one or more processors and comprising:
assembling data indicative of a first query and data identifying one or more external data sources into a first prompt; processing the first prompt using one or more generative models to generate first output data, wherein the first output data includes first source code that is executable to perform an analytic task on data from one or more of the external data sources; causing the first source code to be executed to perform the analytic task using one or more of the external data sources and generate analytic output; assembling the analytic output into a second prompt with a command to determine whether the analytic output satisfies the first query; and processing the second prompt using one or more of the generative models to generate second output data that indicates whether the analytic output satisfies the first query.

20 . The method of claim 19 , wherein the second output data comprises a command to generate second source code that comprises an updat

Assignees

• Google Llc

Inventors

• Papir Alan
• Shtacher Idan Heimlich
• Zhang Tianren
• Chen Jing
• Sobell Samuel
• Potharaju Srividya Pranavi
• Song Yang
• Li Chenmei

Classifications

• G06F16/248

  Presentation of query results · CPC title

• G06F8/71

  Version control (security arrangements therefor G06F21/57); Configuration management · CPC title

• G06F16/242 Primary

  Query formulation · CPC title

• G06F8/43 Primary

  Checking; Contextual analysis · CPC title