# US 2026/0227966 A1 · AI-Powered Iterative Human-in-the-Loop Feedback System

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

- Number: US 2026/0227966 A1
- Document ID: US20260227966A1
- Status: Published application
- Kind: A1
- Application number: 19/358,277
- Filed: 2025-10-14
- Published: 2026-08-06
- Earliest claimed: 2025-02-06
- Assignees: CAST AI Group, Inc.
- Applicants: CAST AI Group, Inc.
- Inventors: Leonid Kuperman
- Web page: https://patentgrep.com/patent/US20260227966A1

## CPC

- G06F 8/35: model driven
- G06F 8/77: Software metrics
- G06F 9/4856: {resumption being on a different machine, e.g. task migration, virtual machine migration (G06F9/5088 takes precedence)}
- G06F 11/3409: {for performance assessment}
- G06F 11/3612: {by runtime analysis (performance monitoring G06F11/3466)}
- G06F 16/284: {Relational databases}
- G06F 11/3419: {by assessing time}
- G06F 2209/482: Application
- H04L 41/5054: Automatic deployment of services triggered by the service manager, e.g. service implementation by automatic configuration of network components
- H04L 43/0817: by checking functioning
- H04L 51/214: using selective forwarding
- H04L 63/029: {Firewall traversal, e.g. tunnelling or, creating pinholes}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)

## Abstract

A system and method for optimizing application performance using user preference feedback. An application performance automation (APA) system receives a proposed modification to source code or configuration files generated by a large language model (LLM), presents the modification to a client device, and receives a natural language comment from the user indicating acceptance or rejection. The system extracts user preferences from the comment and stores them with an association to a user identifier, code repository, or category of performance inefficiency. During a future optimization process, the system detects a candidate modification, retrieves stored preferences, and generates an updated LLM prompt based on the candidate modification and the preferences. The updated prompt is transmitted to the LLM to produce a revised modification tailored to the preferences, which is then presented to the user.

## Claims

### Claim 1

A method for automated application change customization based on persistent user feedback, comprising:
receiving, by an application performance automation (APA) system, a proposed modification to one or more source code files or configuration files of an application, the proposed modification generated by a large language model (LLM);
presenting the proposed modification to a client device of a user;
receiving, from the client device, a user comment in natural language in association with an acceptance or rejection of the proposed modification;
extracting, from the user comment, one or more user preferences;
storing the extracted user preferences in association with at least one of a user identifier, a code repository, or a category of performance inefficiency;
detecting, during a future application performance optimization process, a candidate modification to a code file or configuration file;
retrieving stored user preferences associated with at least one of the user identifier, the code repository, or the category of performance inefficiency;
generating an updated LLM prompt based on the candidate modification and the retrieved user preferences;
transmitting the updated LLM prompt to the LLM to cause the LLM to generate a revised modification that is tailored to the retrieved user preferences; and
presenting the revised modification to the client device of the user.

### Claim 2 (depends on 1)

The method of claim 1, wherein extracting the one or more user preferences from the user comment comprises processing the user comment using a natural language processing (NLP) model.

### Claim 3 (depends on 1)

The method of claim 1, wherein storing the extracted user preferences comprises storing the preferences in a database within a customer-managed computing environment.

### Claim 4 (depends on 1)

The method of claim 1, wherein the one or more user preferences include one or more of a preferred programming language version, a preferred dependency version, or a preferred configuration parameter value.

### Claim 5 (depends on 1)

The method of claim 1, wherein generating the updated LLM prompt comprises inserting the retrieved user preferences into a predefined prompt schema having labeled sections for task description, reasoning, proposed solutions, and an implementation plan.

### Claim 6 (depends on 1)

The method of claim 1, wherein detecting the candidate modification during the future optimization process comprises analyzing telemetry data from an observability platform to identify a performance inefficiency.

### Claim 7 (depends on 1)

The method of claim 1, wherein the revised modification generated by the LLM is presented to the client device as a pull request in a version control system.

### Claim 8 (depends on 1)

The method of claim 1, further comprising, in response to presentation of the revised modification to the client device, receiving a second user comment, updating the stored user preferences based on the second comment, and using the updated preferences in a subsequent modification generation.

### Claim 9 (depends on 1)

The method of claim 1, wherein transmitting the updated LLM prompt to the LLM comprises selecting the LLM from a plurality of available LLMs based on one or more of task complexity, execution context, or data sensitivity.

### Claim 10 (depends on 9)

The method of claim 9, wherein selecting the LLM comprises dynamically choosing between a GPU-deployed LLM and a CPU-deployed LLM based on computational requirements of the candidate modification.

### Claim 11 (depends on 1)

The method of claim 1, wherein retrieving the stored user preferences comprises filtering preferences based on a category of performance inefficiency selected from latency reduction, resource consumption reduction, or throughput improvement.

### Claim 12 (depends on 1)

The method of claim 1, wherein the APA system validates the revised modification in a sandbox environment prior to presenting the revised modification to the client device.

### Claim 13 (depends on 12)

The method of claim 12, wherein validating the revised modification in the sandbox environment comprises running unit tests, integration tests, or performance benchmarks.

### Claim 14 (depends on 1)

The method of claim 1, wherein storing the extracted user preferences comprises associating each stored preference with metadata including a time stamp, originating application identifier, and change history reference.

### Claim 15

A non-transitory computer-readable medium, encoded thereon computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform steps comprising:
receiving, by an application performance automation (APA) system, a proposed modification to one or more source code files or configuration files of an application, the proposed modification generated by a large language model (LLM);
presenting the proposed modification to a client device of a user;
receiving, from the client device, a user comment in natural language in association with an acceptance or rejection of the proposed modification;
extracting, from the user comment, one or more user preferences;
storing the extracted user preferences in association with at least one of a user identifier, a code repository, or a category of performance inefficiency;
detecting, during a future application performance optimization process, a candidate modification to a code file or configuration file;
retrieving stored user preferences associated with at least one of the user identifier, the code repository, or the category of performance inefficiency;
generating an updated LLM prompt based on the candidate modification and the retrieved user preferences;
transmitting the updated LLM prompt to the LLM to cause the LLM to generate a revised modification that is tailored to the retrieved user preferences; and
presenting the revised modification to the client device of the user.

### Claim 16 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein extracting the one or more user preferences from the user comment comprises processing the user comment using a natural language processing (NLP) model.

### Claim 17 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein storing the extracted user preferences comprises storing the preferences in a database within a customer-managed computing environment.

### Claim 18 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the one or more user preferences include one or more of a preferred programming language version, a preferred dependency version, or a preferred configuration parameter value.

### Claim 19 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein generating the updated LLM prompt comprises inserting the retrieved user preferences into a predefined prompt schema having labeled sections for task description, reasoning, proposed solutions, and an implementation plan.

### Claim 20

A computing system, comprising:
one or more processors; and
a non-transitory computer-readable medium, encoded thereon computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform steps comprising:
receiving, by an application performance automation (APA) system, a proposed modification to one or more source code files or configuration files of an application, the proposed modification generated by a large language model (LLM);
presenting the proposed modification to a client device of a user;
receiving, from the client device, a user comment in natural language in association with an acceptance or rejection of the proposed modification;
extracting, from the user comment, one or more user preferences;
storing the extracted user preferences in association with at least one of a user identifier, a code repository, or a category of performance inefficiency;
detecting, during a future application performance optimization process, a candidate modification to a code file or configuration file;
retrieving stored user preferences associated with at least one of the user identifier, the code repository, or the category of performance inefficiency;
generating an updated LLM prompt based on the candidate modification and the retrieved user preferences;
transmitting the updated LLM prompt to the LLM to cause the LLM to generate a revised modification that is tailored to the retrieved user preferences; and
presenting the revised modification to the client device of the user.
