# US 2025/0094143 A1 · Dynamic user interface generation using large language models

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

- Number: US 2025/0094143 A1
- Document ID: US20250094143A1
- Status: Published application
- Kind: A1
- Application number: 18/891,390
- Filed: 2024-09-20
- Published: 2025-03-20
- Earliest claimed: 2023-09-20
- Assignees: Airbnb, Inc.
- Applicants: Airbnb, Inc.
- Inventors: Joseph Huang
- Web page: https://patentgrep.com/patent/US20250094143A1

## CPC

- G06F 8/38: for implementing user interfaces
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

An embodiment includes training a first large language model (LLM) to generate source code implementing an input set of user interface (UI) components, the training resulting in a trained UI generation model. An embodiment includes generating, from a first decision query, a first UI generation task, the first UI generation task comprising a decision criterion and a dataset. An embodiment includes generating, from the first UI generation task, using the trained UI generation model, first source code implementing a first arrangement of UI components. An embodiment includes executing, using a webpage rendering framework, the first source code, the executing rendering the first arrangement of UI components onto a webpage.

## Claims

### Claim 1

A computer-implemented method comprising:
training a first large language model (LLM) to generate source code implementing an input set of user interface (UI) components, the training resulting in a trained UI generation model;
generating, from a first decision query, a first UI generation task, the first UI generation task comprising a decision criterion and a dataset;
generating, from the first UI generation task, using the trained UI generation model, first source code implementing a first arrangement of UI components; and
executing, using a webpage rendering framework, the first source code, the executing rendering the first arrangement of UI components onto a webpage.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein the training comprises fine-tuning, using a dataset of user interface component data, a foundational source code generation LLM.

### Claim 3 (depends on 1)

The computer-implemented method of claim 1, wherein the first UI generation task is generated by prompting a second LLM with the first decision query.

### Claim 4 (depends on 3)

The computer-implemented method of claim 3, wherein the second LLM is a foundational language generation LLM.

### Claim 5 (depends on 1)

The computer-implemented method of claim 1, further comprising:
generating, from the first UI generation task, using the trained UI generation model and feedback corresponding to the first arrangement of UI components, second source code implementing a second arrangement of UI components; and
executing, using the webpage rendering framework, the second source code, the executing rendering the second arrangement of UI components onto the webpage.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, further comprising:
further training, using feedback corresponding to the first arrangement of UI components, the trained UI generation model.

### Claim 7 (depends on 1)

The computer-implemented method of claim 1, further comprising:
storing, in a UI component dataset, stored source code implementing a UI component in the first arrangement of UI components.

### Claim 8 (depends on 7)

The computer-implemented method of claim 7, further comprising:
generating, from a second decision query, a second UI generation task, the second UI generation task comprising a second decision criterion and a second dataset; and
generating, from the second UI generation task, using the trained UI generation model, source code implementing a third arrangement of UI components, the source code comprising the stored source code.

### Claim 9

A non-transitory computer-readable medium storing a program, which when executed by a computer, configures the computer to:
train a first large language model (LLM) to generate source code implementing an input set of user interface (UI) components, the training resulting in a trained UI generation model;
generate, from a first decision query, a first UI generation task, the first UI generation task comprising a decision criterion and a dataset;
generate, from the first UI generation task, using the trained UI generation model, first source code implementing a first arrangement of UI components; and
execute, using a webpage rendering framework, the first source code, the executing rendering the first arrangement of UI components onto a webpage.

### Claim 10 (depends on 9)

The non-transitory computer-readable medium of claim 9, wherein the training comprises fine-tuning, using a dataset of user interface component data, a foundational source code generation LLM.

### Claim 11 (depends on 9)

The non-transitory computer-readable medium of claim 9, wherein the first UI generation task is generated by prompting a second LLM with the first decision query.

### Claim 12 (depends on 11)

The non-transitory computer-readable medium of claim 11, wherein the second LLM is a foundational language generation LLM.

### Claim 13 (depends on 9)

The non-transitory computer-readable medium of claim 9, wherein the program, when executed by the computer, further configures the computer to:
generate, from the first UI generation task, using the trained UI generation model and feedback corresponding to the first arrangement of UI components, second source code implementing a second arrangement of UI components; and
execute, using the webpage rendering framework, the second source code, the executing rendering the second arrangement of UI components onto the webpage.

### Claim 14 (depends on 9)

The non-transitory computer-readable medium of claim 9, wherein the program, when executed by the computer, further configures the computer to:
further train, using feedback corresponding to the first arrangement of UI components, the trained UI generation model.

### Claim 15 (depends on 9)

The non-transitory computer-readable medium of claim 9, wherein the program, when executed by the computer, further configures the computer to:
store, in a UI component dataset, stored source code implementing a UI component in the first arrangement of UI components.

### Claim 16 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the program, when executed by the computer, further configures the computer to:
generate, from a second decision query, a second UI generation task, the second UI generation task comprising a second decision criterion and a second dataset; and
generate, from the second UI generation task, using the trained UI generation model, source code implementing a third arrangement of UI components, the source code comprising the stored source code.

### Claim 17

A system comprising:
a processor; and
a non-transitory computer readable medium storing a set of instructions, which when executed by the processor, configure the system to:
train a first large language model (LLM) to generate source code implementing an input set of user interface (UI) components, the training resulting in a trained UI generation model;
generate, from a first decision query, a first UI generation task, the first UI generation task comprising a decision criterion and a dataset;
generate, from the first UI generation task, using the trained UI generation model, first source code implementing a first arrangement of UI components; and
execute, using a webpage rendering framework, the first source code, the executing rendering the first arrangement of UI components onto a webpage.

### Claim 18 (depends on 17)

The system of claim 17, wherein the training comprises fine-tuning, using a dataset of user interface component data, a foundational source code generation LLM.

### Claim 19 (depends on 17)

The system of claim 17, wherein the first UI generation task is generated by prompting a second LLM with the first decision query.

### Claim 20 (depends on 19)

The system of claim 19, wherein the second LLM is a foundational language generation LLM.
