Chef-GPT is an Azure Function-powered application that uses AI to generate healthy, easy-to-cook recipes, complete with step-by-step instructions and AI-generated images of the final meal. The project leverages GPT-4o mini for recipe generation and DALL-E 3 for creating meal visuals. The entire application is provisioned and deployed using Azure Developer CLI (azd), with model integration managed via Azure AI Studio.
- Healthy Recipe Generation: Create recipes based on user preferences for quick and nutritious meals.
- Step-by-Step Instructions: Provide detailed cooking instructions to guide users through meal preparation.
- AI-Generated Meal Images: Use DALL-E 3 to visualize the final dish based on the recipe.
- Azure Functions: Utilize a serverless architecture to ensure scalability.
- Azure AI Studio Integration: Manage AI models for recipe and image generation.
- Azure Account: An active Azure subscription is required.
- Azure Command-Line CLI (az): Ensure az is installed.
- Azure Developer CLI (azd): Ensure azd is installed.
- Azure AI Studio Models:
- GPT-4o mini for text-based recipe generation.
- DALL-E 3 for image-based meal visualization.
- Ensure both models are created and configured in Azure AI Studio.
Log into the Azure CLI and set the subscription. If on Windows, ensure Azure PowerShell is connected.
az login
azd auth login # Optional: --use-device-code-
Create an azd environment
azd init -e dev
-
Configure the environment variables
| Variable | Purpose | Example |
|---|---|---|
| AZURE_SUBSCRIPTION_ID | The Azure Subscription to deploy to | 29a3884f-d905-4009-8e3f-d41dcfac5e1f |
| AZURE_LOCATION | The Azure Region | eastus |
| GPT_ENDPOINT | The endpoint for the GPT model used for recipe generation | https://<azure-ai-service-name.openai>.azure.com/openai/deployments/gpt-4o-mini/chat/completions?api-version=2024-08-01-preview |
| DALLE_ENDPOINT | The endpoint for the DALL-E model used for image generation | https://<azure-ai-service-name.openai>.openai.azure.com/ |
| AZURE_AI_API_KEY | The Azure AI API key |
```bash
azd env set AZURE_SUBSCRIPTION_ID <Azure subscription id>
azd env set AZURE_LOCATION <Azure Region>
azd env set GPT_ENDPOINT <GPT Endpoint>
azd env set DALLE_ENDPOINT <DALL-E Endpoint>
azd env set AZURE_AI_API_KEY <API Key>
```
azd provisionWhen run locally, the Azure Function runs as the Visual Studio user. Ensure the user has the following roles on the listed resources:
| Resource | Role | Notes |
|---|---|---|
| App Config | App Configuration Data Reader | Note: This role can take up to 15 min to take effect. |
| Key Vault | Key Vault Secrets User |
Create a local.settings.json file in Infrastructure with the following content
{
"IsEncrypted": false,
"Values": {
"AzureWebJobsStorage": "UseDevelopmentStorage=true",
"FUNCTIONS_WORKER_RUNTIME": "dotnet-isolated"
}
}
Because the local.settings.json may contain secrets, such as connection strings, you should never store it in a remote repository. Read more at Code and test Azure Functions locally
The following environment variables must be set to run the function locally. They can be obtained from azd's .env file, or from the resource in Azure.
- APP_CONFIGURATION_ENDPOINT
Start the Infrasturcture project from Visual Studio. Once running, the application can be accessed via an HTTP endpoint provided by the Azure Function. Send a POST request with the desired recipe parameters (e.g., dietary preferences, time constraints) to generate a recipe and receive an AI-generated image of the meal.
Use the demo REST script to test the functionality. You'll need to install the REST Client Visual Studio Extension.
The workspace is brought online using the Azure Developer CLI. Additionally, Visual Studio tasks can be used.
| Action | Command |
|---|---|
| Start | azd up |
| Stop | azd down --purge --force |
| Deploy infra only | azd provision |
| Deploy function | azd deploy |
The model's instructions are stored in gpt-instructions-safety. A Powershell script is provided to easily turn the markdown into a string that can be passed as the model's system message. The safety parameters are adapted from the sample safety instructions in Azure AI Studio, they are not complete and should not be assumed to be correct. The system message is stored in configuration.bicep.
sequenceDiagram
participant User as User
participant FunctionApp as RecipeFunction (Azure Function)
participant Handler as GetRecipeQueryHandler
participant GptService as IGptService
participant ImageService as IImageGenerationService
participant SessionStorage as ISessionStorage
participant HttpClient as HttpClient
participant GPT4OMini as GPT-4o Mini (Azure AI)
participant DALL_E as DALL-E (Azure AI)
User->>FunctionApp: HTTP POST /Prompt (optional session ID)
FunctionApp->>FunctionApp: Parse Request Body
alt Invalid Request
FunctionApp->>User: 400 Bad Request
else Valid Request
FunctionApp->>Handler: Handle GetRecipeQuery
Handler->>GptService: Send recipe prompt
GptService->>SessionStorage: Save User Message with Session ID
SessionStorage-->>GptService: Chat History
GptService->>HttpClient: HTTP POST Request with System Prompt and Optional Chat History to Azure AI Service
HttpClient->>GPT4OMini: Send Prompt
GPT4OMini-->>HttpClient: GPT Response
HttpClient-->>GptService: GPT Response
GptService->>SessionStorage: Save GPT Response with Session ID
GptService-->>Handler: Recipe Instructions
alt Final Recipe Generated
Handler->>ImageService: Generate Image
ImageService->>DALL_E: Request Image Generation
DALL_E-->>ImageService: Image URL
ImageService-->>Handler: Image URL
Handler->>FunctionApp: RecipeResponse with Image URL
else Partial Recipe Generated
Handler->>FunctionApp: RecipeResponse without Image
end
FunctionApp->>User: 200 OK with Recipe Response (includes Session ID)
end
Contributions are welcome! Please submit a pull request or open an issue to suggest improvements.
This project is licensed under the MIT License. See the LICENSE file for details.