.Net: Sample Demo Code Showcasing Usage of Reasoning Models

[joslat]

Motivation and Context

1. Required: showing usage of reasoning effort.
2. Problem: Controlling reasoning effort - no sample
3. Scenario: Using reasoning effort to benefit from the new amazing models :)

Description

This pull request adds sample demo code that demonstrates how to leverage reasoning models by using a AgentBase class that implements this support nicely.
The new implementation allows agents to specify a reasoning effort level (Low, Medium, High) that is applied to the OpenAIPromptExecutionSettings when invoking Semantic Kernel.

The PR includes three demo agents:

• SmartBlogPostAgent (Low Reasoning): Generates an analytical blog post on Semantic Kernel.
• PoemAgent (Medium Reasoning): Composes a creative poem about Semantic Kernel.
• CodeExampleAgent (High Reasoning): Provides a sample C# code snippet demonstrating Semantic Kernel integration.

Additionally, a comprehensive README.md has been added with configuration instructions (using User Secrets or environment variables), guidance on setting the project as the default startup, and how to run the demo.

This implementation not only enhances our agentic AI capabilities but also complements the phenomenal code from Roger Barreto (@RogerBarreto), further strengthening Semantic Kernel overall solution.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

[joslat]

@RogerBarreto I saw there was no sample, so "giving back" a bit ;)
Also, what do you think of the example use cases? Would you change/adjust them to properly use the "reasoning effort"?

[RogerBarreto]

Thanks for the sample and contribution @joslat.
I'm tagging @crickman as one of our Agent expert to have a look in the demo and comment. As we are approaching GA on the agents would be interesting to have your optinion and changes that bring a good Agent with latest reasoning capabilities to our Demo collections.

[joslat]

@RogerBarreto & @crickman any feedback? :)
If not, I think it fits the quality gate so we could approve, after all it's an example on "how to use the new reasoning effort parameter", that's all ;)

[crickman]

@joslat, I suspect reasoning effort might be easily showin in a concepts sample instead of a dedicated demo. Also, this may be accomplished without introducing an agent abstraction that is entirely separate from the agent framework.

Another option might be to use the Agent Framework ChatCompletionAgent only or just use a chat-completion approach.

I am concerned that AgentBase may create some confusion for anyone who might stumble upon this sample.

[joslat]

Hi @crickman
I agree with the removal of AgentBase and simplification of the example.
The demo imho is necesary as otherwise how do we ensure that a reasoning model is being used? here we need an o1 or o3-mini model that supports this.

And i am using the ChatCompletionAgent - inside the AgentBase ;) - thus not that visible.

I'll remove the AgentBase and put the implementatioion on the Agent methods... I was just trying to reduce repeating code...

In regards do making this a concepts sample, is there a way to ensure an o1 or o3-mini model/deployment is used? otherwise I would suggest to keep this as a demo

How do the examples look for you? (blog post for light reasoning, poem for medium and some code for high reasoning effort)

[crickman]

You might consider adding a different configuration parameter for the reasoning model so it can be defined separate from ChatDeploymentName when structured as a focused sample.

I might also not include the hard validation on the model name since it may evolve and azure model deployments can be named anything.

[crickman]

This may be prefered in the current mode:

FunctionChoiceBehavior = FunctionChoiceBehavior.Auto()

[crickman]

Thanks for the sample and contribution @joslat. I'm tagging @crickman as one of our Agent expert to have a look in the demo and comment. As we are approaching GA on the agents would be interesting to have your optinion and changes that bring a good Agent with latest reasoning capabilities to our Demo collections.

@RogerBarreto - The core of the demo isn't really agent specific. I've recommended to @joslat that it be refactored as a focused sample.