AIOps solutions heavily rely on comprehensive knowledge for automating various IT operations such as resource provisioning, service configurations, and patch management. With ignio’s Flamingo release, we are introducing a GenAI-powered feature, ‘Code Accelerator’, to accelerate the process of creation of automation knowledge.

The traditional approach of creating automation knowledge is extensively manual where Subject Matter Experts (SMEs) write code for various operations. This process is time-consuming, the quality of the code varies from person to person, and the generated code is often prone to errors and not the first-time-right. We introduce a paradigm shift in this space by incorporating GenAI to automatically create code based on user instructions. The result is a process that is fast, has consistent quality, and is most often the first-time-right. This enables AIOps solutions to easily adapt to technological changes and accelerate the automation of lifecycle-specific service operations.

Code Accelerator has the following three features.

  • Code Generator: A GenAI app trained to take natural language instructions and generate automation
  • Code Translator: A GenAI app that translates code from one language to another.
  • Code Logic Generator: A GenAI app that generates a natural language description of the code logic.

 

 

Code Accelerator Main Principles

The Code Accelerator design is based on the following three key principles.

  • The first principle is to make it very easy for the user to provide instructions to generate code. A user is required to just mention their instructions in simple English language. For example, “write a code to check the health of an HTTP process”. Alternately, a user is required to just paste a Standard Operating Procedure (SOP) in simple English language. We implement this principle by effectively leveraging Large Language Models (LLMs). LLMs understand natural languages as well as programming languages. Thus, LLMs can act as that bridge where users can mention their instructions or requirements in simple English language for LLMs to generate the equivalent code.
  • The second principle is that Code Accelerator is designed to make code that is customized for ignio. This makes it distinctive from any other code-generating apps. Our Code Accelerator understands the data structures in which ignio knowledge processes input and output. It also understands how pre-processing and post-processing is implemented in ignio. Furthermore, the Code Accelerator understands the semantics of iWorkflow, a custom-built Domain-specific Language (DSL). This has been made possible by using various levers of in-context learning such as few-shot prompting, chain-of-thought prompting, and prompt chaining.
  • The third principle is that Code Accelerator is not just using LLMs to generate code. Additional checks have been incorporated to evaluate the correctness and quality of code using backend code engine and evaluation LLMs.

Code Accelerator follows Responsible AI, a framework of guidelines and principles for developing and using AI in ethical, safe, and responsible manner. Guardrails have been placed to avoid any harmful use, to avoid processing of any PII data, and to prevent hallucinations and jailbreak scenarios.

 

 

Code Generator

Code Generator transforms user instructions provided in natural language into executable code using LLMs. It is very easy to use.

  • User specifies a brief description of what the code should perform – for example, “write a code to check the health of all RHEL hosts managed by a Cloudera manager.”
  • User specifies the language in which the user wants the code. Currently, supported scripting languages are Bash, PowerShell, Groovy, and PL/SQL.
  • A user can also specify if the script should consider any input parameters.

That’s it! The app then generates the code. If the user wants to make any modifications after reading the code, the user can give further instructions, and the app fine-tunes the code as per the instructions.

Code Accelerator uses Prompt Engineering to absorb user inputs and generates the code. Prompts are designed to ensure that the generated code meets the user’s requirements. Code Generator uses Chain-of-Thought (CoT) prompting to generate the code. Initially, an LLM generates the first version of the code. This code is then refined through multiple iterations with the LLM to fix errors, handle inputs correctly, and ensure proper execution, resulting in the final code.

The code generation process also leverages existing ignio knowledge, such as pre-written functions for various automations. This allows users to create complex automation scripts that might involve multiple pre-existing functions within the organization.

Let’s explore some important points to note about Code Generation.

  • This is the first-of-its-kind custom-built GenAI solution that understands ignio semantics. We are working on incorporating further ignio customizations such as performing look-up in the existing OOB functions and performing code validations and optimizations.
  • Code Generator goes beyond generating the code. After generating the code, it performs validations to ensure the quality and correctness of the code.

Code Translator

In the ever-evolving world of IT operations, code translation is a common but often time-consuming and redundant task. For instance, AIOps automation engineers often face a situation where they have developed a code in Bash, and they want an equivalent code in PowerShell. This has been made possible with just a click with Code Translator! Just upload your code developed in one language and select the target language you want an output into, that’s it! Code Translator is particularly useful for translating an existing code to a more optimized language or when an environment no longer supports an older language. Currently, supported scripting language translations are between Bash and PowerShell, and from Groovy to iWorkflow.

There are two types of translations possible with the Code Translator: (1) from a universally known language to another universally known language, and (2) from a universally known language to a custom language.

When making translations between two universally known languages, both programming languages are already known to the LLM. Currently, Code Translator supports translation between two such languages: Bash and PowerShell. Both languages, Bash, and PowerShell, are very specific to their respective operating systems, that is Bash for Linux and PowerShell for Windows. Due to this, some functionalities in one language might not have a direct equivalent to the other. To address these challenges, prompt engineering is used to ensure that the code is translated correctly, whenever possible.

When translating from a universally known language to a custom language, one of the programming languages is not known to the LLM. Currently, Code Translator supports the translation from Groovy to iWorkflow. iWorkflow is a DSL optimized for automation scripts to be run on ignio. This type of translation is implemented using in-context learning and few-shot prompting. The necessary knowledge about iWorkflow is provided to the LLM, and the Groovy code is passed through the LLM repeatedly to ensure that the correct iWorkflow code is generated.

Code Logic Generator

Many times, we discover a code, and we are curious to learn what tasks that code performs. Code Logic Generator does exactly help you there. Code Logic Generator takes your code as an input and produces a clear, step-by-step logical flow. It offers both a brief overview and a detailed breakdown, including steps and sub-steps of the code. Code Logic Generator leverages in-context learning and few-shot prompting techniques to generate the logical flow for the code provided as an input. Currently, Code Logic Generator supports Bash, PowerShell, Groovy, PL/SQL, and iWorkflow languages.

Code Accelerator Key highlights

The following are the key highlights of Code Accelerator.

  • It is a GenAI-powered feature that takes natural language instructions and generates code.
  • It is the first-of-its-kind GenAI solution that understands ignio semantics.
  • It is the first-of-its-kind GenAI solution that understands the iWorkflow semantics and can automatically generate iWorkflow code.
  • It processes user instructions provided in simple English language to generate code. On the other hand, it processes code and generates its explanation in simple English language. Thus, Code Accelerator makes it easy for the users to create, modify, and understand automation knowledge.
  • It is not just generating code, but also assessing and ensuring the quality and the correctness of code.
  • Moreover, it is equipped with Responsible AI guardrails to prevent hallucinations and avoid any harmful usage.

Comparison with other code accelerator and generation tools

What makes Code Accelerator different from other code generation tools available in the market?

Some features that stand out are its ability to generate ignio-specific code, its ability to understand iWorkflow semantics, and its flexibility to implement the latest and the most accurate AI models. The following table presents detailed comparison.

 

 

Conclusion

Digitate’s Code Accelerator reduces the time to implement a new use case to a few hours (if not minutes), reduces dependency on SMEs, and accelerates the time to value. Code Accelerator is our first attempt towards redefining the experience of creating automation knowledge. We have an exciting roadmap ahead to further transform this automation knowledge creation process. Stay tuned!