How to use structured output

[nicocomumu]

JSON output formatting has become widespread, avoiding manual and approximate deserialization.

In Python, I can use the with_structured_output() function as shown in the following example.

How can I do the equivalent in .NET?

# Pydantic définition
class ActionResponse(BaseModel):
    title: str = Field(alias="Title", description="The title of the action.")
    location: str = Field(alias="Location", description="The location name of the action.")
    from_day: datetime = Field(alias="FromDay", description="The day of this action starts.")
    to_day: datetime = Field(alias="ToDay", description="The day of this action ends.")

...

# Sample usage of structured output
    llm = ChatOpenAI(
        model=settings.OPENAI_MODEL_NAME,
        temperature=settings.OPENAI_TEMPERATURE,
        openai_api_key=settings.OPENAI_API_KEY,
    )

    # This is the incredible part !
    structured_llm = llm.with_structured_output(ActionResponse)
	    
    user_message = "sample prompt"

    prompt_template = ChatPromptTemplate.from_messages(
        [
            SystemMessage(content="The persistent system prompt"),
            HumanMessagePromptTemplate.from_template("{user_message}")
        ]
    )

    runnable = (
        {"user_message": RunnablePassthrough()} | prompt_template | structured_llm 
    )

    return runnable.invoke(user_message)

[HavenDV]

Not implemented yet, I recommend using https://devblogs.microsoft.com/dotnet/introducing-microsoft-extensions-ai-preview/ and Semantic Kernel if that's not enough. This project will be more of an extension for this in the future than a separate ecosystem