# US 2025/0094137 A1 · Visual Programming Platform Featuring Machine Learning for Automated Code Development

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0094137 A1
- Document ID: US20250094137A1
- Status: Published application
- Kind: A1
- Application number: 18/468,025
- Filed: 2023-09-15
- Published: 2025-03-20
- Earliest claimed: 2023-09-15
- Assignees: Google Inc.
- Applicants: Google LLC
- Inventors: Ruofei Du; Zhongyi Zhou
- Web page: https://patentgrep.com/patent/US20250094137A1

## CPC

- G06F 8/34: Graphical or visual programming
- G06F 8/33: Intelligent editors
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A visual programming platform can leverage a machine learning-based coding system to generate an initial set of programming-language code for further graphical editing by a human user. As an example, the visual programming platform can obtain a natural language description of a task to be performed by a computational pipeline. The visual programming platform can process the natural language description of the task with a machine learning coding system that includes one or more machine-learned models to generate, as an output of the machine learning coding system, a set of pseudocode that describes performance of the task. The platform can process the set of pseudocode that describes performance of the task with a compiler to generate a set of programming-language code that defines the computational pipeline for performing the task. The visual programming platform can generate a graphical visualization of the computational pipeline defined by the set of programming-language code.

## Claims

### Claim 1

A visual programming platform that enables the graphical development of software code, the visual programming platform implemented by one or more computing devices and configured to perform operations, the operations comprising:
obtaining a natural language description of a task to be performed by a computational pipeline;
processing the natural language description of the task with a machine learning coding system comprising one or more machine-learned models to generate, as an output of the machine learning coding system, a set of pseudocode that describes performance of the task;
processing the set of pseudocode that describes performance of the task with a compiler to generate a set of programming-language code that defines the computational pipeline for performing the task;
generating a graphical visualization of the computational pipeline defined by the set of programming-language code; and
providing the graphical visualization of the computational pipeline for display in an interactive user interface that enables a human user to edit the graphical visualization to modify the set of programming-language code.

### Claim 2 (depends on 1)

The visual programming platform of claim 1, wherein the machine learning coding system comprises one or more large language models.

### Claim 3 (depends on 1)

The visual programming platform of claim 1, wherein the machine learning coding system comprises a machine-learned pseudocode drafting model configured to process the natural language description of the task to generate the set of pseudocode that describes performance of the task.

### Claim 4 (depends on 3)

The visual programming platform of claim 3, wherein the machine learning coding system comprises a machine-learned node selection model configured to process the natural language description of the task to select one or more selected nodes from a library of nodes, wherein the one or more selected nodes are identified to the machine-learned pseudocode drafting model.

### Claim 5 (depends on 4)

The visual programming platform of claim 4, wherein the one or more selected nodes are identified in a prompt supplied to the machine-learned pseudocode drafting model.

### Claim 6 (depends on 5)

The visual programming platform of claim 5, wherein the prompt comprises a few-shot prompt that provides one or more examples of pseudocode output.

### Claim 7 (depends on 1)

The visual programming platform of claim 1, wherein the natural language description of the task to be performed by the computational pipeline is input by the human user.

### Claim 8 (depends on 1)

The visual programming platform of claim 1, wherein the computational pipeline comprises a machine learning pipeline that executes one or more machine learning models to process a pipeline input to generate a pipeline output.

### Claim 9 (depends on 1)

The visual programming platform of claim 1, wherein the graphical visualization of the computational pipeline comprises an editable node-graph.

### Claim 10

A computer-implemented method for finetuning a sequence processing model to perform a pseudocode drafting task, the method comprising:
obtaining, by a computing system comprising one or more computing devices, a training tuple, the training tuple comprising a natural language description of a computational pipeline and a set of programming-language code that defines the computational pipeline;
decompiling, by the computing system, the set of programming-language code to generate a set of ground truth pseudocode;
processing, by the computing system, the natural language description of the computational pipeline with the sequence processing model to generate a set of predicted pseudocode;
evaluating, by the computing system, a loss function that compares the set of predicted pseudocode with the set of ground truth pseudocode; and
modifying, by the computing system, one or more values of one or more parameters of the sequence processing model based on the loss function.

### Claim 11 (depends on 10)

The computer-implemented method of claim 10, wherein the natural language description was generated by a human.

### Claim 12 (depends on 10)

The computer-implemented method of claim 10, wherein the natural language description was generated by a machine learning model.

### Claim 13 (depends on 10)

The computer-implemented method of claim 10, wherein decompiling, by the computing system, the set of programming-language code to generate a set of ground truth pseudocode comprises applying, by the computing system, a set of inverse compiling rules to the set of programming-language code.

### Claim 14 (depends on 10)

The computer-implemented method of claim 10, wherein the loss function evaluates whether the set of predicted pseudocode contains the same number, type, or sequence of nodes as the set of ground truth pseudocode.

### Claim 15

One or more non-transitory computer-readable media that store computer-executable instructions that, when executed, cause a computing system to perform operations, the operations comprising:
obtaining a natural language description of a task to be performed by a computational pipeline;
processing the natural language description of the task with a machine learning coding system comprising one or more machine-learned models to generate, as an output of the machine learning coding system, a set of pseudocode that describes performance of the task;
processing the set of pseudocode that describes performance of the task with a compiler to generate a set of programming-language code that defines the computational pipeline for performing the task;
generating a graphical visualization of the computational pipeline defined by the set of programming-language code; and
providing the graphical visualization of the computational pipeline for display in an interactive user interface that enables a human user to edit the graphical visualization to modify the set of programming-language code.

### Claim 16 (depends on 15)

The one or more non-transitory computer-readable media of claim 15, wherein the machine learning coding system comprises one or more large language models.

### Claim 17 (depends on 15)

The one or more non-transitory computer-readable media of claim 15, wherein the machine learning coding system comprises a machine-learned pseudocode drafting model configured to process the natural language description of the task to generate the set of pseudocode that describes performance of the task.

### Claim 18 (depends on 17)

The one or more non-transitory computer-readable media of claim 17, wherein the machine learning coding system comprises a machine-learned node selection model configured to process the natural language description of the task to select one or more selected nodes from a library of nodes, wherein the one or more selected nodes are identified to the machine-learned pseudocode drafting model.

### Claim 19 (depends on 18)

The one or more non-transitory computer-readable media of claim 18, wherein the one or more selected nodes are identified in a prompt supplied to the machine-learned pseudocode drafting model.

### Claim 20 (depends on 19)

The one or more non-transitory computer-readable media of claim 19, wherein the prompt comprises a few-shot prompt that provides one or more examples of pseudocode output.
