Session

When AI needs to make a move - decision models in action

Choosing a tool, prioritizing a support ticket, or deciding whether to take an action are common tasks in AI applications. Decision models such as Jev, Microsoft Decision-1, and the Polish-developed Basal allow us to describe a situation, ask a question, and get a result from a predefined set of options, along with their probabilities. The application can then use that result to determine its next step.

In this session, I'll explain how decision models work and when to use them instead of rule-based systems, traditional ML models, or generative LLMs. Using routing and prioritization examples, I'll show how to prepare input data, define possible outcomes, and evaluate whether a model can be trusted to make a particular decision.

The second half of the session will be dedicated to a live demo. Microsoft Decision-1, running in Microsoft Foundry, will take control of an application where the environment continues to change even while the model is generating a response. We'll examine the observations sent to the model, the available actions, their probabilities, and what the application ultimately executes.

We'll explore situations where the model gets stuck, lacks the necessary context, or returns a response after the environment has already changed. I'll demonstrate how changes to observations and instructions affect the model's behavior, and how the application verifies whether a proposed action is still valid. Through these examples, we'll learn to distinguish a poor model decision from problems with input data or action execution.

Maciej Kępa

Data & AI Architect at Datumo | MLOps & AI-ready Data Platforms

Kraków, Poland

Actions

Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.

Jump to top