AWS Bedrock Based Utilities Part 1

The integration of advanced AI models, such as Anthropic’ s Claude 3.5 Sonnet, into cloud environments like AWS Bedrock offers transformative capabilities for software development processes. This blog series explores how these technologies can enhance code generation, code analysis, and requirement generation.

Requirement Generation

Generating requirements from high-level project descriptions is another area where Claude 3.5 Sonnet shines. By using few-shot learning techniques, we can guide the model to produce structured requirement documents based on initial prompts.

The typical workflow involves:

·???????? Initial Prompting: provide a brief overview of the project goals.

·???????? Structured Output: The model generates Business requirement, User requirement, Product requirement, technical requirement and Quality & Compliance requirement.

Code Generation

The code generation capabilities of Claude 3.5 Sonnet are particularly noteworthy. We can leverage its advanced natural language processing to automate the creation of code snippets across various programming languages.

This process typically involves:

·???????? Input Specification: Provide natural language prompts that describe the desired functionality.

·???????? Model Invocation: The model is invoked using the AWS SDK, allowing us to send requests directly to the Claude API.

·???????? Output Retrieval: The generated code is returned, ready for integration into existing projects.

?

Code Analysis

Claude 3.5 Sonnet also excels in code analysis, providing insights into existing codebases to identify bugs, suggest optimizations, and ensure adherence to best practices.

The analysis process generally includes:

·???????? Code Input: input the existing code into the model.

·???????? Feedback Generation: The model analyses the code for potential issues or improvements.

·???????? Recommendations: It outputs suggestions for refactoring or optimizing the code.

This capability is particularly beneficial for maintaining large codebases where manual reviews can be time-consuming and prone to oversight. This can be further enhanced to a agentic framework and integrated with GitHub.

?

Prerequisites

EC2 for doing deploying the streamlit application.

• IAM Roles
• Security Groups
• EC2 instance
• AWS Bedrock Model Access

?

Walkthrough

Let us start by logging into the ec2 instance, creating virtual environment, deploying necessary packages and the streamlit application.? All these are already covered in my previous blogs so I will be directly demonstrating the usage of the utilities developed.

Sample Prompt

??Business Requirements

User Requirements

Product Requirements

Technical Requirement

Quality & Compliance

?

Additional Improvements

The Knowledge base is limited to what the Bedrock model has been trained on. We can improvise it further by embedding a search engine and using the data search engine generates to be fed to the model , this way we can have the model look at more data than what it has been trained on.

?

Code Generation Based on Generative AI.

Sample Prompt

Generated code

We can download the whole code and description as zip by pressing Download as ZIP. All the prompts are saved in Prompts history for us to regenerate if required.

?Currently the code can be generated for Nodejs, Python and Java only.

Additional upgrades

·???????? Support other programming Languages.

·???????? Refined to modify existing code and issues seen in code.

·???????? Implement Agentic framework for code generation and integration with AWS Q for developer.

?Code Analysis

We can Analyse single page of code or Multiple files which should be uploaded after zipping the source as one zip file.

Sample Prompt

Let us Analyse the generated code. This is the code we generated previously.

Press “Analyze code”.

Analysis History is Maintained, and we can download when required. Once the application is stopped this data is lost. This will be additional enhancement to the system.

?Now, let us analyse a full zip code.

Press “Analyze code”.

Wait for this to complete.

The report is ready.

We can Press on “Download Analysis” to download report for all the files as zip file.

Additional Improvements

·???????? Implement Agentic framework for Code analysis.

·???????? Integrate Q developer for code analysis.

Cleaning Up

Delete the EC2 instance which are created for the demo.

Conclusion

In this Blog, we learned how to do Generate requirement, Code and Analyse code using Anthropic Claude LLM model.? In the Next Part, we will demonstrate how to build a UI App from wireframe diagram without writing a single line of code.