# US 2026/0093857 A1 · Llm-generated infrastructure design

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

- Number: US 2026/0093857 A1
- Document ID: US20260093857A1
- Status: Published application
- Kind: A1
- Application number: 18/901,714
- Filed: 2024-09-30
- Published: 2026-04-02
- Earliest claimed: 2024-09-30
- Applicants: Oracle International Corporation
- Inventors: Paul Gregory GREENSTEIN; Zbigniew HOLKA
- Web page: https://patentgrep.com/patent/US20260093857A1

## CPC

- G06F 30/13: Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Systems, methods, and other embodiments associated with LLM-based generation of computing infrastructure are described. In one embodiment, an example method includes accessing infrastructure requirements for compute infrastructure that are in human language. The example method includes translating the infrastructure requirements into a physical infrastructure topology using one or more large language models. The example method includes converting the physical infrastructure topology into an executable deployment specification. And, the example method includes executing the deployment specification to automatically configure a target computer system to have the compute infrastructure described by the infrastructure requirements.

## Claims

### Claim 1

One or more non-transitory computer-readable media that include stored thereon computer-executable instructions that, when executed by at least a processor of a computing system, cause the computing system to:
access infrastructure requirements for compute infrastructure that are in human language;
translate the infrastructure requirements into a physical infrastructure topology using one or more large language models;
convert the physical infrastructure topology into an executable deployment specification; and
execute the deployment specification to automatically configure a target computer system to have the compute infrastructure described by the infrastructure requirements.

### Claim 2 (depends on 1)

The one or more non-transitory computer-readable media of claim 1, wherein the computer-executable instructions to translate the infrastructure requirements into the physical infrastructure topology further cause the computing system to:
translate the infrastructure requirements into a logical infrastructure topology using a first large language model that is trained to generate logical infrastructure topologies; and
translate the infrastructure requirements and the logical infrastructure topology into the physical infrastructure topology using a second large language model that is trained to generate physical infrastructure topologies.

### Claim 3 (depends on 2)

The one or more non-transitory computer-readable media of claim 2, wherein the computer-executable instructions to translate the infrastructure requirements into the physical infrastructure topology further cause the computing system to:
present the logical infrastructure topology as a logical infrastructure topology graph for a first user approval before proceeding to translate the logical infrastructure topology into the physical infrastructure topology; and
present the physical infrastructure topology as a physical infrastructure topology graph for a second user approval before proceeding to convert the physical infrastructure topology into the executable deployment specification.

### Claim 4 (depends on 1)

The one or more non-transitory computer-readable media of claim 1, wherein the executable deployment specification is written in YAML code or HCL code.

### Claim 5 (depends on 1)

The one or more non-transitory computer-readable media of claim 1, wherein the computer-executable instructions further cause the computing system to:
access a training set of associated training logical infrastructure topologies, training physical infrastructure topologies, and training infrastructure requirements for other computing infrastructure;
train one large language model of the one or more large language models to generate the physical infrastructure topologies using the training physical infrastructure topologies, training logical infrastructure topologies, and training infrastructure requirements in the training set.

### Claim 6 (depends on 1)

The one or more non-transitory computer-readable media of claim 1, wherein the computer-executable instructions further cause the computing system to validate the translation using one or more further large language models that are trained to detect errors in infrastructure topologies.

### Claim 7 (depends on 1)

The one or more non-transitory computer-readable media of claim 1, wherein the computer-executable instructions to translate the infrastructure requirements into a physical infrastructure topology further cause the computing system to represent the physical infrastructure topology as a collection of tokens in a graph representation language.

### Claim 8

A computer-implemented method, comprising:
accessing infrastructure requirements for compute infrastructure that are in human language;
translating the infrastructure requirements into a logical infrastructure topology using a first large language model that is trained to generate logical infrastructure topologies;
translating the infrastructure requirements and the logical infrastructure topology into a physical infrastructure topology using a second large language model that is trained to generate physical infrastructure topologies;
converting the physical infrastructure topology into an executable deployment specification; and
executing the deployment specification to automatically configure a target computer system to have the compute infrastructure described by the infrastructure requirements.

### Claim 9 (depends on 8)

The computer-implemented method of claim 8, further comprising:
presenting the logical infrastructure topology as a logical infrastructure topology graph for a first user approval before proceeding to translate the logical infrastructure topology into the physical infrastructure topology; and
presenting the physical infrastructure topology as a physical infrastructure topology graph for a second user approval before proceeding to convert the physical infrastructure topology into the executable deployment specification.

### Claim 10 (depends on 8)

The computer-implemented method of claim 8, wherein the executable deployment specification is written in YAML code.

### Claim 11 (depends on 8)

The computer-implemented method of claim 8, wherein the executable deployment specification is written in HCL code.

### Claim 12 (depends on 8)

The computer-implemented method of claim 8, further comprising:
automatically validating the logical infrastructure topology with a third large language model that is trained to detect whether there exist errors in the logical infrastructure topology; and
automatically validating the physical infrastructure topology with a fourth large language model that is trained to detect whether there exist errors in the physical infrastructure topology.

### Claim 13 (depends on 8)

The computer-implemented method of claim 8, further comprising:
in response to detection of an error in the logical infrastructure topology, automatically update first training data for the first large language model with corrections to the error in the logical infrastructure topology, and re-train the first large language model with the updated first training data; and
in response to detection of an error in the physical infrastructure topology, automatically update second training data for the second large language model with corrections to the error in the physical infrastructure topology, and re-train the second large language model with the updated second training data.

### Claim 14 (depends on 8)

The computer-implemented method of claim 8, wherein the logical infrastructure topology and the physical infrastructure topology are generated as collections of tokens in a graph representation language.

### Claim 15

A computing system, comprising:
a processor;
a memory;
one or more non-transitory computer-readable media that include stored thereon computer-executable instructions that, when executed by at least the processor, cause the computing system to:
access infrastructure requirements for compute infrastructure that are in human language;
translate the infrastructure requirements into a physical infrastructure topology using one or more large language models;
convert the physical infrastructure topology into an executable deployment specification; and
execute the deployment specification to automatically configure infrastructure of the computing system as described by the infrastructure requirements.

### Claim 16 (depends on 15)

The computing system of claim 15, wherein translation of the infrastructure requirements into the physical infrastructure topology further comprises translating the infrastructure requirements into an intermediate logical infrastructure topology prior to producing the physical infrastructure topology.

### Claim 17 (depends on 16)

The computing system of claim 16, wherein the computer-executable instructions to translate the infrastructure requirements into the physical infrastructure topology further cause the computing system to:
present the logical infrastructure topology as a logical infrastructure topology graph for a first user approval before proceeding to translate the logical infrastructure topology into the physical infrastructure topology; and
present the physical infrastructure topology as a physical infrastructure topology graph for a second user approval before proceeding to convert the physical infrastructure topology into the executable deployment specification.

### Claim 18 (depends on 16)

The computing system of claim 16,
wherein the computer-executable instructions to translate the infrastructure requirements into a logical infrastructure topology further cause the computing system to represent the logical infrastructure topology as a first collection of tokens in a graph representation language; and
wherein the computer-executable instructions to translate the infrastructure requirements and the logical infrastructure topology into a physical infrastructure topology further cause the computing system to represent the physical infrastructure topology as a second collection of tokens in the graph representation language.

### Claim 19 (depends on 15)

The computing system of claim 15, wherein the executable deployment specification is written in YAML code or HCL code.

### Claim 20 (depends on 15)

The computing system of claim 15, wherein the computer-executable instructions further cause the computing system to:
access a training set of associated training logical infrastructure topologies, training physical infrastructure topologies, and training infrastructure requirements for other computing infrastructure;
train the one or more large language models to generate the physical infrastructure topologies using the training g infrastructure requirements and at least one of the training physical infrastructure topologies, training logical infrastructure topologies in the training set.
