This is an AI-powered chat application that automates the creation of infrastructure as code (IaC) for Azure infrastructure using Bicep, leveraging LLMs to understand user requirements and generate the code. To make sure the generated code is accurate and adheres to best practices, it incorporates feedback from the Bicep lint tool and retrieves Microsoft docs references. The system operates in the following phases:

1. Understands user requirements interactively through a chat interface.
2. Summarizes the requirements once the system determines enough information has been collected.
3. Generates Bicep code based on the summarized requirements.
4. Validates the generated code using Bicep linting tool.
5. If linting errors are found, it retrieves relevant context from the linting results and Bicep documentation, and regenerates the code.
6. Repeats the generation and validation cycle until the code passes linting or reaches a maximum number of attempts.

• Target: Azure engineers (especially Azure Infra engineers)
• Problem: Time-consuming and error-prone to create Azure environments with unique requirements (e.g. PoC)
• Goal: Streamline and automate the process of generating Bicep code for Azure infrastructure, making it easier and faster for engineers to create tailored environments.

• Python 3.12+
• Node.js 18+
• pnpm
• Azure CLI or bicep CLI

1. Configure the Azure OpenAI environment variables in a .env file.

AZURE_OPENAI_ENDPOINT=https://your-endpoint.openai.azure.com/
AZURE_OPENAI_DEPLOYMENT_NAME=your-deployment-name
OPENAI_API_VERSION=2024-02-15-preview
CHAT_DEPLOYMENT_NAME=your-chat-deployment-name
CODE_DEPLOYMENT_NAME=your-code-deployment-name

2. Start the backend server.

cd backend

# Option 1: venv
python3 -m venv .venv
.venv/bin/activate # Linux/macOS
.venv\Scripts\activate.ps1 # Windows PowerShell
pip install -r requirements.txt
python -m uvicorn app:app --host 127.0.0.1 --port 8000 --reload --app-dir ..

# Option 2: uv
uv run -m uvicorn app:app --host 127.0.0.1 --port 8000 --reload --app-dir ..

3. Start the frontend server.

cd frontend
pnpm install
pnpm dev

4. Access the application at http://localhost:3000.

graph TD
    A((Start)) --> B1{Are requirements sufficient?}
    B1 -- Yes --> C[Summarize requirements]
    B1 -- No --> B2[Ask a question]
    B2 --> B3[User response]
    B3 --> B1

    C -- Use user requirements as context --> D[Generate Bicep code]
    D --> E[Lint the Bicep code]
    E --> F{Are there lint errors?}
    F -- Yes --> G1[Retrieve documentation]
    G1 --> G2[Generate guidance to fix errors]
    G2 -- Use lint errors and guidance as context --> D
    F -- No --> Y[Final Bicep code]
    Y --> Z[User review and download]
    Z --> End((End))

    classDef llm stroke:#333,stroke-width:1px,stroke-dasharray: 4;
    class B1,B2,C,D,G2 llm;

This project is licensed under the MIT License (SPDX: MIT). See the LICENSE file for details.