Using JSON Schema for Structured Output in .NET for OpenAI Models

Using JSON Schema for Structured Output in .NET for OpenAI Models:

• Demonstrated implementing a JSON Schema-based structured output using .NET version of Semantic Kernel for OpenAI models.
• JSON Schema ensures well-structured responses, follows specified format, and easily deserialized by the system.

Supported Models for Structured Outputs:

• Azure OpenAI: Access to or later The API version
• OpenAI model

Structured Outputs with .NET Version of Semantic Kernel:

• Approach 1: Providing JSON Schema to OpenAI model by initializing object and setting it in property
  • Example: Getting top 10 movies of all time from the model and specifying JSON Schema of the desired response
  • Result: Defined JSON Schema ensures consistent response format across all AI model responses
• Approach 2: Defining desired response format using C# class or structure and setting its type directly in .NET
  • Example: Defining response format using C# class or structure and using the same type to deserialize model response
  • Result: Easy access to properties in the code with a predefined response format

Limitations:

• Some keywords from JSON Schema are not supported in OpenAI Structured Outputs yet

Conclusion:

• Using JSON Schema and Semantic Kernel framework allows control over AI-generated response format for structured, predictable, and easy-to-use outputs
• Reach out for questions or feedback through Semantic Kernel GitHub Discussion Channel
• Support Semantic Kernel by giving a star on GitHub if you've enjoyed using it