System and methods for generating compliant infrastructure as code using generative ai
Generative AI and machine-learning techniques automate the generation and validation of infrastructure as code, addressing inefficiencies and errors in manual processes by ensuring compliance and adaptability in distributed computing environments.
Patent Information
- Application Number
- PCT/US2025/027696
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-05-05
- Publication Date
- 2026-11-12
AI Technical Summary
Manual generation and maintenance of infrastructure as code (IaC) to meet enterprise demands is time-consuming and prone to human error, especially in distributed computing environments with varying policies and changing requirements.
Utilizing generative AI and machine-learning techniques to automatically generate and validate infrastructure as code (IaC) by training and tuning models with tuning data sets, ensuring compliance with policy as code (PaC) rules, and continuously monitoring for compliance.
Reduces time and errors in generating compliant IaC, enhances infrastructural stability, and adapts to changing requirements by automating the generation and validation process.
Smart Images

Figure US2025027696_12112026_PF_FP_ABST
Abstract
Description
PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO SYSTEM AND METHODS FOR GENERATING COMPLIANT INFRASTRUCTURE AS CODE USING GENERATIVE AlTechnical Field
[0001] The present disclosure relates generally to artificial intelligence. More specifically, but not by way of limitation, this disclosure relates to systems and methods for generating enterprise compliant infrastructure as code using generative artificial intelligence.Background
[0002] With the growth of distributed computing environments and cloud computing, provisioning of traditional computing infrastructure has grown increasingly virtualized. Virtualized infrastructure can be represented via Infrastructure as Code C'laC ). laC refers to the ability to manage and provision, within the distributed computing environment, computer resources such as databases, virtual machines, operating systems, and the like via code configurations.
[0003] Various enterprises may rely on cloud vendors and laC to build and manage computing environments as required by the respective enterprise. The demands of such enterprises, including internal policies, external policies, and the tasks to be performed by a given distributed computing environment can utilize various configurations of laC. For instance, one enterprise may use laC configured to meet internal policies drastically different from the policies implemented by another enterprise. Moreover, changes in a given enterprise's policies or infrastructure may result in modifications to the laC previously used to instantiate a distributed computing environment. However, manual generation and maintenance of the laC to meet different enterprise demands or to meet changing demands of a given enterprise can be time-consuming and prone to human error.Summary
[0004] Various examples of the present disclosure provide systems and methods for generating compliant infrastructure-as-code for deployment in various distributed computing environments. In one example, a non-transitory computer-readable storage medium having program code executable by a processing can be executed to perform a set of operations. The set of operations include receiving input requirements associated with a distributed computing 1US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO environment, generating a set of infrastructure as code (“IaC”) requirements, and generating a tuning data set including previous IaC code examples based in part on the IaC requirements. The operations can include identifying a selected model based in part on the input requirements and tuning the selected model with the tuning data set to generate a tuned model. The tuned model can generate an IaC output. The operations can include validating the infrastructure as code output based on a comparison of the infrastructure as code output against a set of policy as code (“PaC”) rules. In response to successfully validating the IaC output, the operations can include deploying the IaC output to the distributed computing environment.
[0005] In another example, a method that includes one or more processing devices performing a set of operations is described. The set of operations include receiving input requirements associated with a distributed computing environment, generating a set of infrastructure as code requirements, and generating a tuning data set including previous IaC code examples based in part on the laC requirements. The operations can include identifying a selected model based in part on the input requirements and tuning the selected model with the tuning data set to generate a tuned model. The tuned model can generate an laC output. The operations can include validating the infrastructure as code output based on a comparison of the infrastructure as code output against a set of PaC rules. In response to successfully validating the laC output, the operations can include deploying the IaC output to the distributed computing environment.
[0006] In yet another example, a computing system is described. The computing system includes a processing device and non-transitory computer-readable storage medium having program code executable by the processing device to perform a set operations. The set of operations include receiving input requirements associated with a distributed computing environment, generating a set of infrastructure as code requirements, and generating a tuning data set including previous IaC code examples based in part on the laC requirements. The operations can include identifying a selected model based in part on the input requirements and tuning the selected model with the tuning data set to generate a tuned model. The tuned model can generate an laC output. The operations can include validating the infrastructure as code output based on a comparison of the infrastructure as code output against a set of PaC rules. In2US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO response to successfully validating the laC output, the operations can include deploying the laC output to the distributed computing environment.
[0007] This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used in isolation to determine the scope of the claimed subject matter. The subject matter should be understood by reference to appropriate portions of the entire specification, any or all drawings, and each claim.
[0008] The foregoing, together with other features and examples, will become more apparent upon referring to the following specification, claims, and accompanying drawings.Brief Description of the Drawings
[0009] FIG. 1 is a block diagram depicting an example of a computing environment in which an infrastructure as code (“laC”) generation computing system can generate compliant laC, according to certain aspects of the present disclosure.
[0010] FIG. 2 is a flow diagram depicting an example of a process generating compliant laC, according to certain aspects of the present disclosure.
[0011] FIG. 3 is a flow diagram depicting an example of a process generating, deploying, and monitoring laC, according to certain aspects of the present disclosure.
[0012] FIG. 4 is a block diagram depicting an example of a computing device, which can be used to implement the laC generation server or the model tuning server, according to certain aspects of the present disclosure.Detailed Description
[0013] Certain aspects and features of the present disclosure involve generating compliant infrastructure as code (“laC’’) using generative Al techniques. The generated laC may then be applied to various distributed computing environments including cloud computing environments to deploy computing architecture which is compliant with various policies associated with the entity operating the distributed computing environment. Generative Al and machine-learning techniques can facilitate the deployment and management of the computing services deployed per the compliant laC. The use of the described techniques can address gaps3US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO in techniques for generating compliant laC, thereby improving efficiency, compliance, and adaptability to changing requirements.
[0014] To generate the compliant laC, the described examples can include techniques for training and tuning generative Al models such as large language models (“LLMs”) for generating the compliant laC. To do so, the systems and methods can generate tuning data used for tuning the generative Al models, referred to more generally as machine-learning models throughout. Generally, the operations described here can relate to tuning a pre-built model such as a pre-trained LLM. Similar techniques may be applied to build and train a machine-learning model. Thus, reference to tuning data and model tuning can similarly refer to training data and model training. Generating tuning data can include defining specific laC requirements including compute, network, resource, and compliance requirements and retrieving compliant laC based on the laC requirements. The laC requirements may be received, for instance, via a user-interface where a user provides information related to the distributed computing environment in which the laC is to be deployed. The information can include various policies that the distributed computing environment is to comply with and other technical requirements of the distributed computing environment. In some examples, the user can provide direct examples of previously compliant laC or non-compliant laC, or a combination of both, which can serve as training samples.
[0015] Using the training samples, provided via the user interface, or retrieved based on requirements input into the user interface, the described examples can further include preparing tuning data that includes positive or negative examples of laC, or a combination of both, where positive examples include laC complying with the defined requirements, and negative examples include laC identified as failing to comply with the defined requirements. In some examples, reinforcement learning techniques are applied in which each of the positive and negative laC examples are used to train or tune the machine-models for generating the compliant laC. In other examples, other tuning techniques can be used such that only positive or negative laC examples are used.
[0016] Preexisting machine-learning models for generating compliant laC can be selected based on determined characteristics of the distributed environment for integration. Examples of machine-learning models selected can include Generative Pre-trained Transformer (“GPT”)-3, Pathways Language model (“PaLM”), Language Model for Dialogue Applications 4US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO (“LaMDA ’), OpenAI Codex, and the like. Such models, previously trained on various corpora, may nevertheless lack context dependent capabilities to generate compliant laC. Therefore, once selected for inclusion in a production environment, the models may be tuned per the tuning datasets which are curated for laC compliance, including examples of compliant or non-compliant laC, or a combination of both. According to some examples, reinforcement learning procedures may be further used to train and tune the selected machine-learning model to enhance compliance. Once trained and tuned, the selected machine-learning model may then be seamlessly integrated into existing laC pipeline where compliant laC is generated.
[0017] Validation processes may be implemented prior to the deployment of generated laC to ensure compliance of any laC deployed distributed computing network. Policy as code C’PaC") systems may be used, for instance, to continuously check the generated laC. If such validations fail, processes and interfaces for addressing and correcting non-compliant laC can be applied according to some examples. Similarly, techniques for providing feedback to the machine-learning model to further tune the machine-learning model for re-generating compliant laC are described. Validation processes may also be implemented subsequent to deployment, such as through continuous validation procedures to further ensure compliance, for instance, in response to updated compliance requirements or changes in computing architecture. Validation processes implemented subsequent to deployment, according to some examples, include monitoring the distributed computing environment. Beyond evaluating the laC architecture for compliance with structural requirements, performance requirements such as maximum latency, minimum bandwidth, and the like may be continuously monitored and evaluated. In response to determining the performance is below a given threshold, the system can automatically retune the machine-learning model and generate new laC for deployment, compliant both with PaC rules and network performance requirements.
[0018] Certain aspects described herein provide improvements to the infrastructural stability of distributed computing environments. For instance, the described techniques includes accessing requirements and restrictions related to distributed computing environment compute types, network configurations, and resource limits to improve the governing architecture within the distributed computing environment. Such processes may be performed automatically, bypassing traditional requirements for the manual generation of laC. Manual laC generation and compliance validation may be time consuming and prone to human error.5US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO Instead, machine-learning models may be tuned based on tuning data corresponding to computing requirements and restrictions. As a result, automatic generation of compliant laC not only reduces time consuming efforts, but furthers the infrastructural stability of the distributed computing environment by reducing error. Moreover, techniques are described for continuous validation and generation of laC to ensure compliance after integration of machinelearning models or deployment of laC code. For instance, in some examples, in response to detecting violations of determined requirements, the systems and methods can either automatically retrain the machine-learning model, deploy new laC to automatically resolve distributed computing environment non-compliance and performance issues, or both retrain and the machine-learning model and deploy new laC.
[0019] These illustrative examples are given to introduce the reader to the general subject matter discussed here and are not intended to limit the scope of the disclosed concepts. The following sections describe various additional features and examples with reference to the drawings in which like numerals indicate like elements, and directional descriptions are used to describe the illustrative examples but, like the illustrative examples, should not be used to limit the present disclosure.Operating Environment Example for laC generation
[0020] Referring now to the drawings, FIG. 1 is a block diagram depicting an example of an operating environment 100 in which an laC generation computing system 130 builds and trains a machine-learning model that can be used to generate compliant laC, according to certain aspects of the present disclosure. FIG. 1 depicts examples of hardware components of an laC generation computing system 130. The laC generation computing system 130 is a specialized computing system that may be used for processing large amounts of data using a large number of computer processing cycles. The number of devices depicted in FIG. 1 are provided for illustrative purposes. Different numbers of devices may be used. For example, while certain devices or systems are shown as single devices in FIG. 1, multiple devices may instead be used to implement these devices or systems.
[0021] The laC generation computing system 130 can include a model tuning server 108 for tuning one or more machine-learning models (e.g., building and training or otherwise retuning large language models and other models for generating laC). The laC generation computing system 130 can include an laC generation server 106 for generating compliant 6US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO infrastructure-as-code based on requirements 112 which may be applied to machine-learning model(s) 118. Specifically, the laC generation server 106 can host an laC generation application 110. where the laC generation application 110 can receive requirements 112, which may be received via a client computing system 102 across a public data network 104 or may be retrieved across an internal data network 120 from a data repository 122. The requirements 112 may be communicated to the model tuning server 108 and model tuning application 116 which may then train or tune the machine-learning model(s) 118.
[0022] Tuning (or training) the machine-learning model(s) 118 can be performed on the model tuning server 108 via the model tuning application 116 by generating tuning samples 128. The training samples can include requirements 124 stored in a data repository 122 or received via the client computing system 102. Tuning samples 128 can similarly include various versions 126 of infrastructure-as-code. For example, either positive examples (e.g.. previously determined compliant IaC),negative examples (determined non-compliant laC), or a combination of both, may be included within the tuning samples 128 according to various examples. In some examples, one or more of positive examples of compliant laC versions 126 and negative examples of non-compliant laC versions 126 may be used as part of the tuning samples 128 as called by the model tuning application 1 16 to train or tune machine-learning model(s) 118 for generating compliant laC code.
[0023] Each of the laC generation server 106 and the model tuning server 108 can include one or more processing devices that execute program code, such as the laC generation application 110 or the model tuning application 116. The program code is stored on a non-transitory computer-readable medium (e g., as discussed with respect to FIG. 4). The model tuning application 116 can execute one or more processes to train or tune machine-learning model(s) 118 for generating laC, while the laC generation application 110 can execute one or more processes to interface with client computing systems 102 and the model tuning server 108 as discussed above.
[0024] The laC generation computing system 130 can communicate with various other computing systems such as client computing systems 102. For example, the laC generation computing system 130 may include one or more provider external -facing devices that communicate with data provider systems for receiving the data regarding entities or other items to be stored in records in the data repository 122.7US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO
[0025] The client computing systems 102 may interact, via one or more public data networks 104, with various external-facing subsystems of the laC generation computing system 130. For instance, an individual can use a client computing system 102 to submit requests for the creation of compliant laC for deployment to a given distributed computing environment. Within the request, the user can provide the requirements 124 for the distributed computing environment to be built, which may be transmitted via the public data network 104 through the client external -facing subsystem 132 to the laC generation server 106. The client externalfacing subsystem 132 can selectively prevent the client computing system 102 from accessing or searching in the data repository7122. For example, the client external-facing subsystem 132 can determine whether the client computing system 102 can submit requests for generation of distributed computing environment architecture via the laC generation server 106.
[0026] Each client computing system 102 may include one or more third-party devices, such as individual servers or groups of servers operating in a distributed manner. Client computing system 102 can include any computing device or group of computing devices operated by an enterprise user, developer, system manager, auditor, or any other user seeking to access laC including the generation of new laC or previous versions of implemented laC 126 stored in a data repository 122. Client computing system 102 can include one or more server devices. The one or more server devices can include or can otherwise access one or more non-transitory computer-readable media. The client computing system 102 can also execute an online service. The online service can include executable instructions stored in one or more non-transitory computer-readable media.
[0027] Each communication within or with the laC generation computing system 130 may occur over one or more data networks, such as the public data network 104, internal data network 120, or some combination thereof. A data network may include one or more of a variety of different types of networks, including a wireless network, a wired network, or a combination of a wired and wireless network. Examples of suitable networks include the Internet, a personal area network, a local area network (“LAN”), a wide area network (“WAN”), or a wireless local area network (“WLAN”). A wireless network may include a wireless interface or a combination of wireless interfaces. A wired network may include a wired interface. The wired or wireless networks may be implemented using routers, access points, bridges, gateways, or the like, to connect devices in the data netw ork.8US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO
[0028] A data network may include network computers, sensors, databases, or other devices that may transmit or otherwise provide data to the laC generation computing system 130. For example, a data network may include local area network devices, such as routers, hubs, switches, or other computer networking devices. The data networks depicted in FIG. 1 can be incorporated entirely within (or can include) an intranet, an extranet, or a combination thereof. In one example, communications between two or more systems or devices can be achieved by a secure communications protocol, such as secure Hypertext Transfer Protocol (‘’HTTPS”) communications that use secure sockets layer (“SSL”) or transport layer security (“TLS”). In addition, data or transactional details communicated among the various computing devices may be encry pted. For example, data may be encrypted in transit and at rest.
[0029] laC generation server 106 can implement performance monitoring for analyzing and reacting the health of a given distributed computing environment with deployed laC. For instance, the performance monitor 134 can operate with the model tuning server 108 to generate machine-learning models for generating PaC, or can configure PaC per non-machine-leaming techniques. The generated PaC may be used to validate laC generated by other machinelearning model(s) 118. The performance monitor 134 can include configurable rules which detect violations of PaC, preventing the deployment of laC determined to be invalid. Similarly, the performance monitor 134 can include network monitoring tools such as internet control message protocol (“ICMP”) Ping tests, bandwidth monitoring tests, and the like which may monitor the health and performance of the distributed computing environment within which any laC generated code is deployed.
[0030] The laC generation computing system 130 may also include one or more network-attached storage units on which various repositories, databases, or other data structures are stored. Examples of these data structures are the data repository 122. Network-attached storage units may store a variety of different types of data organized in a variety of different ways and from a variety of different sources. For example, the network-attached storage unit may include storage other than the primary storage located within the laC generation server 106 or the model tuning server 108 that is directly accessible by processors located therein. In some aspects, the network-attached storage unit may include secondary, tertiary, or auxiliary storage, such as large hard drives, servers, virtual memory, among other types. Storage devices may include portable or non-portable storage devices, optical storage devices, and various other mediums 9US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO capable of storing and containing data. A machine-readable storage medium or computer-readable storage medium may include a non-transitory medium in which data can be stored and that does not include earner waves or transitory electronic signals. Examples of a non-transitory medium may include, for example, a magnetic disk or tape, optical storage media such as compact disk or digital versatile disk, flash memory, memory, or memory' devices.
[0031] In some aspects, the laC generation computing system 130 can implement one or more procedures to secure communications between the laC generation computing system 130 and other client systems. Non-limiting examples of features provided to protect data and transmissions between the laC generation computing system 130 and other client systems include secure web pages, encryption, firewall protection, network behavior analysis, intrusion detection, etc. In some aspects, transmissions with client systems can be encrypted using public-key cryptography algorithms using a minimum key size of 130 bits. In additional or alternative aspects, website pages or other data can be delivered through HTTPS, secure filetransfer protocol (“SFTP”), or other secure server communications protocols. In additional or alternative aspects, electronic communications can be transmitted using Secure Sockets Layer C'SSL”) technology or other suitable secure protocols. Extended Validation SSL certificates can be utilized to clearly identify a website’s organization identity. In another non-limiting example, physical, electronic, and procedural measures can be utilized to safeguard data from unauthorized access and disclosure.
[0032] The number of devices depicted in FIG. 1 is provided for illustrative purposes. Different numbers of devices may be used. For example, while certain devices or systems are shown as single devices in FIG. 1, multiple devices may instead be used to implement these devices or systems. Similarly, devices or systems that are shown as separate, such as the laC generation server 106 and the model tuning server 108 may be instead implemented in a signal device or system.Examples of Compliant Infrastructure as Code Generation Operations
[0033] FIG. 2 is a flow diagram depicting an example of a process generating compliant laC, according to certain aspects of the present disclosure. For illustrative purposes, the process 200 is described with reference to implementations described above with respect to one or more examples described herein. Other implementations, however, are possible. In some aspects, the operations in FIG. 2 may be implemented in program code that is executed by one or more 10US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO computing devices such as the laC generation server 106 or model tuning server 108 depicted in FIG. 1. In some aspects of the present disclosure, one or more operations shown in FIG. 2 may be omitted or performed in a different order. Similarly, additional operations not shown in FIG. 2 may be performed.
[0034] At block 202, the process 200 involves receiving input requirements associated with a distributed computing environment. Input requirements generally refer to requirements and restrictions which the generated infrastructure must adhere to. Examples of such input requirements can include specific compute, network, and resource requirements and restrictions. For instance, a given input requirement may specify that particular database structures (e.g.. a flat file database, relational database or the like) or file formats are to be used within a deployed distributed computing environment, or that the deployed distributed computing environment is not to exceed a specified storage capacity. The input requirements may be received via a user interface. Additionally or alternatively, input requirements may be retrieved based on prior identified requirements or based on preexisting data stored within the laC generation computing system 130. For instance, the laC generation server 106 can scan a distributed computing environment (e.g., via web-agents and crawlers) to determine some or all of the requirements within the distributed computing environment. In the same or other examples, the laC generation server 106 can communicate with the data repository 122 to identify' requirements 124 of the distributed computing environment based on partial information provided via the user interface.
[0035] At block 204, the process 200 involves generating a set of laC requirements and a tuning data set including previous laC code examples based in part on the set of laC requirements. The set of laC requirements may directly correspond to the input requirements. For instance, the set of infrastructure code requirements may comprise the input requirements received per block 202. Additionally or alternatively, the set of laC requirements may be determined by a performance monitor 134 configured to scan a specified distributed computing environment. For example, the input requirements can specify the type of distributed computing environment that potential laC is to be deployed in. In response, the performance monitor 134 can scan the distributed computing environment directly or can query a database containing requirements associated with the distributed computing environment to identify the set of laC requirements. Examples of laC requirements, as discussed further with respect to 11US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO block 304 of the example process 300, can include computing requirements, network configuration requirements, resource limit requirements, or the like.
[0036] The tuning data set generated per block 204 can include previous laC examples which may be based in part on the laC requirements. For instance, the tuning data can include either positive example laC (e.g., a set of previous laC code examples identified as compliant with the laC requirements), negative example laC (e.g., a set of previous laC code examples identified as non-compliant with the laC requirements), or a combination of both positive and negative examples of laC. As discussed further with respect to the operations of FIG. 3, synthetic laC files that violate requirements can be included within the generate tuning data.
[0037] In some examples, the tuning data can include data other than laC examples. For instance, the tuning data can include documentation, policies and guidelines which describe laC requirements associated with the entity. Entity metadata associated with the specific entity providing input via the user interface can be used to identify relevant documents related to the entity such that tuning data can automatically be added, according to some examples. In the same or other examples, the entity, through the user interface can provide direct copies of the documentation, policies, guidelines related to the laC requirements, and the like.
[0038] At block 206, the process 200 involves identifying a selected model based in part on the input requirements. Examples of models can include LLM machine-models such as Generative Pre-trained Transformer (“GPT”)-3, Pathways Language model (‘'PaLM”), Language Model for Dialogue Applications (“LaMDA”), OpenAI Codex, and the like. Models used can also include code-specific models which may designed and trained specifically for the tasks of laC generation. The machine-learning models may be selected based on the specific needs of laC generation.
[0039] At block 208, the process 200 involves tuning the selected model with the tuning data set to generate a tuned model. The selected model may be further tuned by applying various reinforcement learning procedures to reward compliant code generation and penalize non-compliant code. Algorithms such as Proximal Policy Optimization (“PPO”), Q-leaming, Monte Carlo Tree Search (“MCTS”), Tree-Structured Parzen Estimators (“TPE”) can be used to further tune the model. Reinforcement learning procedures can include either positively weighting the compliant laC, negatively weighting the non-compliant laC, or both positively12US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO weighting compliant laC and negatively weighing non-compliant laC as included within the tuning data generated per block 208.
[0040] At block 210, the process 200 involves generating, by the tuned model, an laC output. Block 210 can include integrating the machine-learning model tuned per block 208 into the infrastructure pipeline, for instance via an API into an laC development workflow. The laC output generated by the machine-learning model tuned per block 208 may, on a first pass, be compliant with the requirements defined per blocks 202 and 204, depending on the accuracy of the tuned model. However, process 200 describes further operations for ensuring the compliance of the generated laC output.
[0041] At block 212, the process 200 involves validating the laC output based on a comparison of the laC output against a set of PaC rules. PaC rules may be retrieved from various repositories (e.g., data repository 122). via the performance monitor 134, or via another system or application communicatively coupled to the laC generation server 106. Additionally or alternatively, the performance monitor 134 can validate the laC output based on the distributed computing environment subsequent to the laC output being deployed within the distributed computing environment (e.g., subsequent to block 212). In such a manner, the laC output may be continuously validated, including in instances prior to deployment at block 212. and subsequent to deployment per block 212.
[0042] At block 214, the process 200 involves, in response to successfully validating the laC output, deploy ing the laC output into the distributed computing environment. Deploy ing the laC output can include using the now validated laC output to configure and deploy infrastructure resources (e.g., compliant databases, load balancers, operating systems, and the like) within the distributed computing environment. As discussed per block 212, the performance monitor 134 can continuously monitor the distributed computing environment subsequent to deploy ment by evaluating not only the continued compliance against PaC rules or any updated requirements of the distributed computing environment, but also in validating the distributed computing environment against several benchmarks related to network performance such as network bandwidth and latency.Examples of Compliant Infrastructure as Code Generation Operations13US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO
[0043] FIG. 3 is a flow diagram depicting an example of a process generating, deploying, and monitoring laC. according to certain aspects of the present disclosure. For illustrative purposes, the process 300 is described with reference to implementations described above with respect to one or more examples described herein. Other implementations, however, are possible. For example, many of the operations described with respect to FIG. 3 may overlap with those described with respect to FIG. 2. In some examples, the operations described in each of FIGs. 2 and 3 may be combined to form further examples still. In some aspects, the operations in FIG. 3 may be implemented in program code that is executed by one or more computing devices such as the laC generation server 106 or model tuning server 108 depicted in FIG. 1. In some aspects of the present disclosure, one or more operations shown in FIG. 3 may be omitted or performed in a different order. Similarly, additional operations not shown in FIG. 3 may be performed.
[0044] At block 302, the process 300 involves receiving a receiving user input from a user interface. The user interface can include any device capable of providing output to a user and receiving input from the user, such as a computer, laptop, mobile device, and the like. The user interface can allow users to specify new or updated compliance requirements, submit tuning data, and also allow users to retrieve and review generated laC. In some examples, each version of laC generated per process 300 may be stored in a repository for access via the user interface such that users can select and compare versions of generated laC for review, to facilitate review of various versions and to provide compliance and logging of the generated laC. The received user input specifically can include receiving input requirements as discussed with respect to block 202 of FIG. 2.
[0045] At block 304, the process 300 involves defining laC requirements. The specific laC requirements can be based in part on the received input requirements per block 302. Additionally, requirements can be based on restrictions that a given machine-learning model for generating the laC output code must adhere to. Such requirements may be identified automatically via the laC generation server 106, or can be identified based on the provided user input. Examples of requirements and restrictions can include compute ty pes used to instantiate the distributed computing environment (e.g., nl -standard- 1, nl-highmem-2, and the like). Requirements and restrictions can include network configurations such as defined permissible subnets and security' groups, firewall rules, and any other specific network configurations 14US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO required for different resource types. Requirements and restrictions can further include resource limits which limit the number of instances able to be run in the deployed distributed computing environment, disks, load balancers within a given laC output, and can specify the types of resources allowed.
[0046] At block 306, the process 300 involves preparing tuning data. Tuning data can include positive examples including compliant laC. Compliant laC can be retrieved from databases communicatively coupled to the laC generation server 106. Additionally, compliant laC may be provided via a user through the user interface discussed with respect to block 302. In either case, compliant laC can be gathered to provide a representative sample of existing laC fdes that comply with the requirements defined per block 304. Generating the tuning data can further include compiling documentation retrieved from various databases or received via user input. Documentation can include an entity’s associated policy documentation, external policy documentation (e.g., public compliance requirements), and other restrictions.
[0047] According to some examples, the tuning data can also include negative examples including non-compliant laC. Similar to the compliant laC, the non-compliant laC can be retrieved from databases communicatively coupled to laC generation server 106 or be provided via user input into the user interface. In further examples, the laC generation server 106 can generate synthetic laC files that violate the requirements defined per block 306 (e g., using disallowed compute types, incorrect network configurations, exceeding defined resource limits, or a combination thereof).
[0048] At block 308, the process 300 involves choosing and tuning a machine-learning model. The tuning operations per block 308 are similar to those discussed with respect to block 208 of process 200, where various machine-learning models may be evaluated and selected for tuning based on the laC requirements. The operations can further include tuning the selected model according to the generated tuning data and various reinforcement learning techniques.
[0049] At block 310, the process 300 integrating the chosen model into an laC pipeline. The laC pipeline can automate the deployment and management of the laC defined infrastructure in a given distributed computing environment. Integration of the chosen model into the pipeline can be implemented via API access such that laC is generated w ithin the laC infrastructure pipeline and laC development workflow. In such a way, the selected, tuned model may be seamlessly integrated within a larger network.15US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO
[0050] At block 312, the process 300 involves validating the laC generated by the machinelearning model complies with all defined requirements and restrictions. Validation can include implementation of PaC systems which can continuously validate the generated laC. Additionally or alternatively, custom checks, for instance received via the user interface, can be implemented which provide further assurance that specific policies, including policies unique to the entity or organization, are adhered to. In some examples, compliance scores can be generated which correspond to the generated laC output’s compliance against the PaC used to validate the generated laC. The compliance scores can negatively correlate to the number of policies violated and the severity' of policies violated. More severe policy violations, identified by the PaC, can lead to greater reductions in the compliance scores. In other examples, in place of explicit compliance scores, any single identified violation of the PaC can result in a determined validation failure.
[0051] At decision 314, the process 300 involves determining a subsequent action based on the results of validation at block 312. Per validation at block 312, if the laC generated by the machine-learning model complies with a threshold compliance score or number of policies, then decision 314 will proceed to block 316 where the laC is deployed. Otherwise, if the generated laC is determined not to comply with all defined requirements and restrictions, or otherwise falls under a threshold compliance score, the process 300 proceeds to block 318, where the process includes handling validation failures. The threshold compliance score can be directly correlated to results of the PaC and custom check testing discussed per block 312. In some examples, the compliance score can be a binary score requiring no identified failures per PaC and custom check testing to be considered passing or validated. In other examples, the compliance score can reflect gradations to be compared against the threshold compliance score.
[0052] At block 316, the process 300 involves deploying the laC. The laC can be deployed into the distributed computing environment where requirements were defined per blocks 302 and 304. As a result of decision 314, the laC has been determined to be compliant with the requirements defined on input per blocks 302 and 304. The laC may then be deployed to a production environment to configure and deploy infrastructure resources.
[0053] In some examples, after the infrastructure resources are deployed, the process 300 can include continuously monitoring (e.g., via the performance monitor 134) the deployed infrastructure to ensure it remains compliant and performs as expected. For example, the 16US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO performance monitor 134 can continuously validate the deployed laC against the PaC. In some examples, the performance monitor 134 can monitor the distributed computing environment for infrastructural changes implemented in the distributed computing environment which originate outside the deployment of the laC output. Additionally or alternatively, the performance monitor 134 can perform network health tests on the distributed computing environment such as ICMP Pings, bandwidth monitoring, latency monitoring and the like, and in response to failures (e.g., scoring under a threshold bandwidth or exceeding a threshold latency), proceed to block 318.
[0054] At block 318, the process 300 involves handling validation failures. As a result of decision 314, the laC may be determined to be non-compliant with various requirements, preventing the laC being deployed to a production environment, further preventing the deployment of infrastructure resources. Instead, handling validation failures per block 318 can include addressing and correcting non-compliant laC as described per blocks 320-326. In other examples, where the laC was previously deployed but nevertheless failed subsequent validation per the performance monitor 134, the deployed laC, determined to be non-compliant can similarly proceed to blocks 320-326.
[0055] At block 320, for instance, the process 300 involves generating signals indicating errors and causing error signals to be displayed on various interfaces, such as the user interface which received the user input per block 302. The error signals can include warnings which identify the specific non-compliance issues. The non-compliance issues can include the specific PaC failures identified, the custom check failures identified, specific network performance failures, any other requirements or combination of requirements defined per block 304 which were identified as violated. Further, the warning signals can be logged in a respective error log or repository to provide for versioning, documentation, and for auditing and compliance.
[0056] At block 322, the process 300 involves adjusting and updating tuning data used to tune the laC machine-learning model. Adjusting tuning data can include adjustments responsive to the validation per decision 314. For instance, in creating and adjusting tuning data where the tuning data can include compliant laC and optionally non-compliant laC, the compliant laC can include the laC deployed per block 316. Additionally. laC that was determined invalid per decision 314 can be added to the tuning data within the non-compliant 17US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO data set in examples where the machine-learning model is tuned according to positive and negative examples. In further examples, additional negative examples in the form of synthetic non-compliant laC can be created and added to the tuning data.
[0057] Independent of or in tandem with the adjustments made in response to the decision 314, adjustment and updating of the tuning data can include manually input updates received via the user interface. At block 324, the process 300 is shown to include receiving updated compliance requirements. Block 324 is similar to the processes of described at block 304, where the process 300 includes establishing laC specific policies, requirements and restrictions that the machine-learning model for generating laC must comply with. However, block 324 is shown as modifying a machine-learning model that has already been integrated into a production pipeline. Thus, users, providing input per user interface at block 302 can initialize policy changes and other changes to requirements used to produce the tuning data and retrain or retune the machine-learning model, and can also provide continuous updates and feedback to machine-learning models which are already deployed.
[0058] At block 326. the process 300 includes retuning or retraining the machine-learning model which has previously been integrated into the pipeline per 310. Block 326 performs operations similar to block 308, where tuning data may be used to further fine tune the machinelearning model and better align the model, now7previously deployed, with the specific requirements identified per blocks 302 and 324. For instance, per determined validation failure at decision 314, updated tuning data can further include additional negative examples of laC non-compliant code previously used to deploy infrastructure. In the same or other examples, the model can be retuned at block 326 where no failures or invalidations occurred at decision 314, and instead, compliant laC is further used to retune the machine-learning model based on additional positive example tuning data. In such a manner the generation of compliant laC may be continuously positively reinforced. In some examples, the retuning per block 326 can occur on a periodic, scheduled basis, or can occur in response to positive or negative validations per decision 314.Example of a Computing System for Machine-Learning Operations
[0059] Any suitable computing system or group of computing systems can be used to perform the operations for the machine-learning operations described herein. For example, FIG. 4 is a block diagram depicting an example of a computing device 400. which can be used 18US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO to implement the laC generation server 106 or the model tuning server 108, according to certain aspects of the present disclosure. The computing device 400 can include various devices for communicating with other devices in the operating environment, as described with respect to FIG. 1. The computing device 400 can include various devices for performing one or more transformation operations described above with respect to FIGS. 1-3.
[0060] The computing device 400 can include a processor 402 that is communicatively coupled to a memory 404. The processor 402 executes computer-executable program code stored in the memory 404, accesses information stored in the memory 404, or both. Program code may include machine-executable instructions that may represent a procedure, a function, a subprogram, a program, a routine, a subroutine, a module, a software package, a class, or any combination of instructions, data structures, or program statements. A code segment may be coupled to another code segment or a hardware circuit by passing or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc. may be passed, forwarded, or transmitted via any suitable means including memory7sharing, message passing, token passing, network transmission, among others.
[0061] Examples of a processor 402 include a microprocessor, an application-specific integrated circuit, a field-programmable gate array, or any other suitable processing device. The processor 402 can include any number of processing devices, including one. The processor 402 can include or communicate with a memory7404. The memory7404 stores program code that, when executed by the processor 402. causes the processor to perform the operations described in this disclosure.
[0062] The memory 404 can include any suitable non-transitory computer-readable medium. The computer-readable medium can include any electronic, optical, magnetic, or other storage device capable of providing a processor w ith computer-readable program code or other program code. Non-limiting examples of a computer-readable medium include a magnetic disk, memory chip, optical storage, flash memory7, storage class memory. ROM, RAM, an ASIC, magnetic storage, or any other medium from which a computer processor can read and execute program code. The program code may include processor-specific program code generated by a compiler or an interpreter from code written in any suitable computerprogramming language. Examples of suitable programming language include Hadoop, C, C++, C#, Visual Basic, Java, Python, Perl, JavaScript, ActionScript, etc.19US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO
[0063] The computing device 400 may also include a number of external or internal devices such as input or output devices. For example, the computing device 400 is shown with an input / output interface 408 that can receive input from input devices or provide output to output devices. A bus 406 can also be included in the computing device 400. The bus 406 can communicatively couple one or more components of the computing device 400.
[0064] The computing device 400 can execute program code 414 that includes the laC generation application 110, the model tuning application 116, or both the laC generation application 110 and the model tuning application 116. The program code 414 for each of the laC generation application 110 and the model tuning application 116 may be resident in any suitable computer-readable medium and may be executed on any suitable processing device. For example, as depicted in FIG. 4, the program code 414 for the laC generation application 110 and the model tuning application 116 can reside in the memory' 404 at the computing device 400 along with the program data 416 associated with the program code 414, such the laC generation application 110 and the machine-learning model(s) 118. Executing the laC generation application 110 or the model tuning application 116 can configure the processor 402 to perform the operations described herein.
[0065] In some aspects, the computing device 400 can include one or more output devices. One example of an output device is the network interface device 410 depicted in FIG. 4. A network interface device 410 can include any device or group of devices suitable for establishing a wired or wireless data connection to one or more data networks described herein. Non-limiting examples of the network interface device 410 include an Ethernet network adapter, a modem, etc.
[0066] Another example of an output device is the presentation device 412 depicted in FIG.4. A presentation device 412 can include any device or group of devices suitable for providing visual, auditory, or other suitable sensory output. Non-limiting examples of the presentation device 412 include a touchscreen, a monitor, a speaker, a separate mobile computing device, etc. In some aspects, the presentation device 412 can include a remote client-computing device that communicates with the computing device 400 using one or more data networks described herein. In other aspects, the presentation device 412 can be omitted.
[0067] In some aspects, dispute scores can be determined and corresponding actions implemented according to one or more of the following examples. As used below, any 20US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO reference to a series of examples is to be understood as a reference to each of those examples disjunctively (e.g., ‘"Examples 1-4" is to be understood as “Examples 1, 2, 3, or 4”).
[0068] Example #1: A method for generating compliant infrastructure as code (“laC”) using generative Al including defining specific compute, network, and resource requirements and restrictions; preparing a training dataset including compliant and non-compliant laC examples; selecting and fine-tuning a GenAI model to align with the defined requirements; integrating the trained GenAI model into an instruction pipeline for laC generation; and continuously validating the generated laC using existing policy as code ('"PaC").
[0069] Example #2: The method of Example #1 , where the GenAI includes large language models (“LLMs”) or code-specific models.
[0070] Example #3: The method according to any of the preceding examples, further including using reinforcement techniques to enhance compliance.
[0071] Example #4: The method according to any of the preceding examples, where the training dataset includes documentation, policies, and guidelines describing the laC requirements and restrictions associated with an entity'.
[0072] Example #5: A system for generating compliant laC including: a module for defining laC requirements and restrictions; a data preparation module for creating a training dataset; a generative artificial intelligence ("GenAI") model fine-tuned to the defined requirements; an integration module for embedding the trained GenAi model into an infrastructure pipeline; and a validation module for ensuring compliance of generated laC.
[0073] Example #6: The system according to Example #5, further including a mechanism for continuous learning and updating of the GenAi model to adapt to evolving compliance requirements.
[0074] Example #7: The system according to any of Examples #5-6, where the validation module interacts with PaC systems to provide feedback on non-compliant laC and fine-tune the GenAi model.
[0075] Example #8: The system according to any of Examples #5-7, where the data preparation module includes the generation of synthetic non-compliant laC examples to enhance the training process.21US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO
[0076] Example #9: The system according to any of Examples #5-8, further including API access for the GenAI model to integrate into existing infrastructure pipelines.
[0077] Example #10: The system according to any of Examples #5-9. where the training dataset includes both positive examples including compliant laC and negative examples including non-compliant laC to train the GenAI model effectively.
[0078] Example #11: The system according to any of Examples #5-10, where the user interface can receive specified requirements and submitted training data, and retrieve generated laC, the user interface further providing tools for monitoring model performance, viewing validation results, and updating compliance requirements.
[0079] Example #12: A method for generating various types of code using GenAI, including the steps outlined according to Example #1, where the types of code include application configuration, security policies, and automation scripts.General Considerations
[0080] Numerous specific details are set forth herein to provide a thorough understanding of the claimed subject matter. However, those skilled in the art will understand that the claimed subject matter may be practiced without these specific details. In other instances, methods, apparatuses, or systems that would be known by one of ordinary' skill have not been described in detail so as not to obscure claimed subject matter.
[0081] Unless specifically stated otherwise, it is appreciated that throughout this specification that terms such as "‘processing,” “computing,” “determining,” and “identifying” or the like refer to actions or processes of a computing device, such as one or more computers or a similar electronic computing device or devices, that manipulate or transform data represented as physical electronic or magnetic quantities within memories, registers, or other information storage devices, transmission devices, or display devices of the computing platform.
[0082] The system or systems discussed herein are not limited to any particular hardware architecture or configuration. A computing device can include any suitable arrangement of components that provides a result conditioned on one or more inputs. Suitable computing devices include multipurpose microprocessor-based computing systems accessing stored 22US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO software that programs or configures the computing system from a general purpose computing apparatus to a specialized computing apparatus implementing one or more aspects of the present subject matter. Any suitable programming, scripting, or other type of language or combinations of languages may be used to implement the teachings contained herein in software to be used in programming or configuring a computing device.
[0083] Aspects of the methods disclosed herein may be performed in the operation of such computing devices. The order of the blocks presented in the examples above can be varied — for example, blocks can be re-ordered, combined, or broken into sub-blocks. Certain blocks or processes can be performed in parallel.
[0084] The use of “adapted to” or “configured to” herein is meant as open and inclusive language that does not foreclose devices adapted to or configured to perform additional tasks or steps. Additionally, the use of “based on” is meant to be open and inclusive, in that a process, step, calculation, or other action “based on” one or more recited conditions or values may, in practice, be based on additional conditions or values beyond those recited. Headings, lists, and numbering included herein are for ease of explanation only and are not meant to be limiting.
[0085] While the present subject matter has been described in detail with respect to specific aspects thereof, it will be appreciated that those skilled in the art, upon attaining an understanding of the foregoing, may readily produce alterations to, variations of, and equivalents to such aspects. Any aspects or examples may be combined with any other aspects or examples. Accordingly, it should be understood that the present disclosure has been presented for purposes of example rather than limitation, and does not preclude inclusion of such modifications, variations, or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art.23US2008310335061
Claims
PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WOClaimsWHAT IS CLAIMED IS:
1. A non-transitory computer-readable storage medium having program code executable by a processing device to perform operations comprising:receiving input requirements associated with a distributed computing environment; generating a set of infrastructure as code (“laC”) requirements and a tuning data set including previous laC code examples based in part on the set of laC requirements;identifying a selected model based in part on the input requirements;tuning the selected model with the tuning data set to generate a tuned model; generate, by the tuned model, an laC output;validate the laC output based on a comparison of the laC output against a set of policy as code C‘PaC’?) rules; andin response to successfully validating the laC output, deploying the laC output into the distributed computing environment.
2. The non-transitory computer-readable storage medium of claim 1, wherein the previous laC code examples include set of positive examples comprising compliant laC and a set of negative examples comprising non-compliant laC.
3. The non-transitory computer-readable storage medium of claim 2, wherein tuning the selected model includes applying a reinforcement learning procedure including positively weighting the compliant laC, and negatively weighting the non-compliant laC.
4. The non-transitory computer-readable storage medium of claim 1, wherein the set of laC requirements comprises, computing requirements, network configuration requirements, resource limit requirements, or any combination thereof.
5. The non-transitory computer-readable storage medium of claim 1, wherein the operations further comprise, subsequent to deploying the laC output, monitoring the laC output based on a comparison of the distributed computing environment against the set of PaC rules.24US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO 6. The non-transitory computer-readable storage medium of claim 5, wherein the operations further comprise, in response to determining the laC output is invalid:identifying the laC output as anon-compliant laC;updating the tuning data set to include the non-compliant laC as a negative example; andre-tuning the tuned model based on the updated tuning data set.
7. The non-transitory computer-readable storage medium of claim 1, wherein the operations further comprise, subsequent to deploying the tuned model, in response to identifying policy changes or infrastructural changes in the distributed computing environment, re-tuning the tuned model.
8. A method that includes one or more processing devices performing operations comprising:receiving input requirements associated with a distributed computing environment; generate a set of infrastructure as code ('‘laC’’) requirements and a tuning data set including previous laC code examples based in part on the set of laC requirements;identify a selected model based in part on the input requirements;tune the selected model with the tuning data set to generate a tuned model; generate, by the tuned model, an laC output;validate the laC output based on a comparison of the laC output against a set of policy as code (“PaC”) rules; andin response to successfully validating the laC output, deploying the laC output into the distributed computing environment.
9. The method of claim 8, wherein generating the tuning data set includes retrieving a set of positive examples comprising compliant laC and retrieving a set of negative examples comprising non-compliant laC.
10. The method of claim 9, wherein tuning the selected model includes applying a reinforcement learning procedure including positively weighting the compliant laC, and negatively weighting the non-compliant laC.25US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO11. The method of claim 8. wherein the set of laC requirements comprises, computing requirements, network configuration requirements,resource limit requirements, or any combination thereof.
12. The method of claim 8, the operations further comprising, subsequent to deploying the laC output, monitoring the laC output based on a comparison of the distributed computing environment against the set of PaC rules.
13. The method of claim 12, the operations further comprising, in response to determining the tuned model is invalid:identifying the laC output as non-compliant laC:updating the tuning data set to include the non-compliant laC as a negative example; andre-tuning the tuned model based on the updated tuning data set.
14. The method of claim 12, the operations further comprising, subsequent to deploying the tuned model, in response to identifying policy changes or infrastructural changes distributed computing environment, re-tuning the tuned model.
15. A computing system comprising:a processing device;a non-transitory computer-readable storage medium having program code executable by the processing device to perform operations comprising:receiving input requirements associated with a distributed computing environment;generate a set of infrastructure as code (“laC’") requirements and a tuning data set including previous laC code examples based in part on the set of laC requirements;identify a selected model based in part on the input requirements; tune the selected model with the tuning data set to generate a tuned model; generate, by the tuned model, an laC output;26US2008310335061PATENT Attorney Docket No 096923-1474406 Client Ref. No. EFX-206WO validate the laC output based on a comparison the laC output against a set of policy as code (“PaC”) rules: andin response to successfully validating the laC output, deploying the laC output into the distributed computing environment.
16. The computing system of claim 15, wherein generating the tuning data set includes retrieving a set of positive examples compnsing compliant laC and retrieving a set of negative examples comprising non-compliant laC.
17. The computing system of claim 16. wherein tuning the selected model includes applying a reinforcement learning procedure including positively weighting the compliant laC, and negatively weighting the non-compliant laC.
18. The computing system of claim 15. wherein the set of laC requirements comprises, computing requirements network configuration requirements, limit requirements, or any combination thereof.
19. The computing system of claim 15, the operations further comprising, subsequent to deploying the laC output, monitoring the laC output based on a comparison of the distributed computing environment against the set of PaC rules.
20. The computing system of claim 19, the operations further comprising, in response to determining the tuned model is invalid:identifying the laC output as non-compliant laC;updating the tuning data set to include the non-compliant laC as a negative example; andre-tuning the tuned model based on the updated tuning data set.27US2008310335061
