# US 2026/0310541 A1 · System debugging

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

- Number: US 2026/0310541 A1
- Document ID: US20260310541A1
- Status: Published application
- Kind: A1
- Application number: 19/715,467
- Filed: 2026-06-22
- Published: 2026-10-08
- Earliest claimed: 2024-01-03
- Assignees: NVIDIA Corporation
- Applicants: NVIDIA Corporation
- Inventors: Sasikumar Neli Sundaram; Gunaseelan Ponnuvel; Chris Siao
- Web page: https://patentgrep.com/patent/US20260310541A1

## CPC

- G06F 11/079: {Root cause analysis, i.e. error or fault diagnosis (in a hardware test environment G06F11/22; in a software test environment G06F11/36)}
- G06F 11/0769: {Readable error formats, e.g. cross-platform generic formats, human understandable formats}
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06F 40/40: Processing or translation of natural language (natural language analysis G06F40/20; semantic analysis G06F40/30)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Disclosed are apparatuses, systems, and techniques that train and use trained language models to assist users with complex systems installation, troubleshooting, and/or maintenance. The techniques include receiving, via a user interface (UI), a natural language (NL) query associated with one or more malfunction indicators indicative of a malfunction state of a system, providing, to a language model (LM) trained using a documentation associated with the system, an input having a prompt that is based at least on the NL query. The techniques further include receiving, from the LM, a response to the NL query, the response having one or more instructions associated with resolution of the malfunction state of the system and causing the UI to display the response.

## Claims

### Claim 1

A method comprising:
providing, to a language model (LM) trained, at least in part, using documentation associated with a system, a prompt that is based at least on a natural language (NL) query associated with one or more malfunction indicators in a test log representative of a malfunction state of the system; and
receiving, from the LM, a response to the NL query, the response including one or more instructions associated with resolution of the malfunction state of the system that includes at least one of:
replacement of a component of the system,
installation of software on the system, or
reinstallation of software on the system; and
causing a user interface (UI) to display the one or more instructions.

### Claim 2 (depends on 1)

The method of claim 1, wherein the one or more malfunction indicators in the test log are generated in association with performed diagnostics of the system.

### Claim 3 (depends on 2)

The method of claim 2, wherein the malfunction state of the system is detected by a diagnostic tool having access to at least one of a hardware of the system or a software operating on the system.

### Claim 4 (depends on 1)

The method of claim 1, further comprising performing an action responsive to the one or more instructions, wherein the resolving the malfunction state of the system comprises:
generating, responsive to the action, an updated test log associated with an updated state of the system;
providing, to the LM, an updated prompt that is based at least on the updated test log;
receiving, from the LM, an updated response to the NL query, the updated response including one or more additional instructions associated with resolution of the updated state of the system.

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
providing, to the LM, an initial input based at least on an initial NL query;
receiving, from the LM, an initial response to the initial NL query, the initial response including one or more initial instructions;
wherein the one or more malfunction indicators are indicative of an initial attempt to resolve the malfunction state of the system.

### Claim 6 (depends on 1)

The method of claim 1, wherein the LM includes a large language model.

### Claim 7 (depends on 1)

The method of claim 1, wherein the documentation associated with the system includes one or more of:
a diagnostic documentation for the system,
a system architecture documentation for the system, or
one or more training diagnostic logs associated with historical or simulated malfunctions of the system.

### Claim 8 (depends on 1)

The method of claim 1, wherein the NL query is generated responsive to at least one of:
a scheduled maintenance of the system,
a hardware update of the system,
a software update of the system,
a firmware update of the system, or
an inoperable condition of the system.

### Claim 9 (depends on 1)

The method of claim 1, wherein the query includes at least one of a text prompt, a speech prompt, an audio prompt, or an image prompt.

### Claim 10 (depends on 1)

The method of claim 1, wherein the prompt is further based at least on one or more embeddings associated with the test log.

### Claim 11

A method comprising:
updating, responsive to an update of a system including at least one of a hardware update, a software update, or a firmware update, training of a language model (LM) using training data including:
a natural language (NL) query associated with one or more malfunction indicators indicative of a malfunction state of the system,
documentation associated with the system, and
a training diagnostic log associated with the malfunction state of the system;
wherein the LM was previously trained to generate an NL response to the NL query, the NL response including one or more instructions associated with resolution of one or more malfunction states of the system prior to the update of the system.

### Claim 12 (depends on 11)

The method of claim 11, wherein the documentation associated with the system includes one or more of:
a diagnostic documentation for the system, or
a system architecture documentation for the system.

### Claim 13 (depends on 11)

The method of claim 11, wherein the training diagnostic log includes at least one of:
a diagnostic log associated with performed diagnostics of the system, or
a synthetic diagnostic log for the malfunction state of the system.

### Claim 14 (depends on 11)

The method of claim 11, wherein the training data further comprises:
a sample NL response to the NL query, and
wherein the updating training of the LM comprises:
applying the training data to the LM to obtain a training NL response; and
causing, using a difference between the training NL response and the sample NL response, one or more parameters of the LM to be modified.

### Claim 15 (depends on 11)

The method of claim 11, wherein the updating training of the LM comprises:
applying the training data to the LM to obtain a training NL response;
obtaining an evaluation score characterizing effectiveness of the training NL response for resolution of the malfunction state of the system; and
causing, using the evaluation score, one or more parameters of the LM to be modified.

### Claim 16 (depends on 11)

The method of claim 11, wherein the LM includes a neural network-based large language model.

### Claim 17 (depends on 11)

The method of claim 11, wherein the training diagnostic log is represented using one or more embeddings generated using a trained embedding generator.

### Claim 18 (depends on 11)

The method of claim 11, wherein the one or more instructions associated with resolution of the one or more malfunction states of the system include at least one of: instructions associated with replacement of a component of the system, or instructions associated with installation of software on the system.

### Claim 19 (depends on 11)

The method of claim 11, further comprising deploying the LM after the updating to provide, responsive to a prompt based on a natural language query associated with one or more malfunction indicators in a test log representative of a malfunction state of the system, one or more instructions associated with resolution of the malfunction state.

### Claim 20

One or more processors comprising processing circuitry to generate and present one or more of software installation instructions or hardware component replacement instructions corresponding to a system, the instructions generated based at least on a language model (LM) processing a natural language prompt associated with one or more malfunction indicators in a test log representative of a malfunction state of the system.
