Nashua CLOUD .NET User Group

Agenda:
6:00 PM - 7:30 PM - AI Workflows with Microsoft Agent Framework

Abstract:
AI systems become reliable when workflows—not prompts—define execution.
In this session, we’ll focus on Microsoft Agent Framework Workflows, the core abstraction for building structured, deterministic, and observable AI systems.

You’ll learn how workflows model AI behavior as explicit execution graphs composed of executors, edges, and type-safe messages, enabling planning, tool execution, reflection, human-in-the-loop interventions, and controlled retries. We’ll explore how orchestrations coordinate these workflows, manage state, and control execution paths—bringing predictability and debuggability to multi-step AI processes.

Rather than relying on opaque agent loops, Microsoft Agent Framework workflows make control flow, decision points, human approvals, and system behavior explicit, making them suitable for production, governance, and enterprise integration.

### Key takeaways:

• Why workflows are the foundational concept in Agent Framework
• Core workflow primitives: executors, edges, messages, checkpoints
• How orchestrations manage flow, state, execution paths, and human-in-the-loop steps
• When to use workflows instead of prompt-driven agents
• Designing AI systems that are deterministic, testable, observable, and governable

In this third session of the Azure Machine Learning series, we’ll focus on the infrastructure that powers your machine learning workloads: compute and environments. After preparing and organizing your data, the next step is understanding how Azure ML runs your code, scales your workloads, and ensures consistent execution across development and production.

This session explores the practical side of working with compute targets and environments in Azure ML. You’ll learn how to select the right compute for your workloads, configure environments for reproducibility, and optimize performance and cost. Whether you’re continuing from Step 2 or already experimenting with training jobs, this session will help you build a solid foundation for running machine learning workloads efficiently in Azure.

You’ll learn:

• How compute and environments fit into the machine learning lifecycle
• The different compute options in Azure ML (compute instances, compute clusters, serverless, and attached compute)
• How to create and manage compute resources using the Studio UI, SDK, and CLI
• How to define and use environments (conda, Docker) for consistent and reproducible runs
• Techniques for selecting the right VM sizes (CPU vs GPU) based on workload needs
• How to manage dependencies and avoid common environment issues
• Best practices for scaling, cost optimization, and auto-scaling compute clusters
• How compute and environments are used in training jobs, pipelines, and deployments

This session is designed to help you move from “I have data ready” to “I can reliably run and scale my ML workloads.” If you’re ready to understand what’s happening behind the scenes and ensure your models run consistently across environments, this is your next step.

In this fourth session of the Azure Machine Learning series, we’ll build on the foundation of clean, well-prepared data and move into the core of machine learning: training models and running experiments. Azure Machine Learning provides powerful capabilities for building, tracking, and optimizing models at scale, enabling teams to move from data to insights efficiently and reproducibly.

This session focuses on the practical side of modeling and experimentation within Azure ML. You’ll learn how to take prepared datasets and turn them into trained models while leveraging Azure ML’s experiment tracking, automation, and scalability features. Whether you’re continuing from Step 3 or already have experience with ML concepts, this session will help you operationalize your modeling workflow in Azure.

You’ll learn:

• How modeling fits into the machine learning lifecycle
• How to create and run training jobs in Azure ML (Studio, SDK, CLI)
• How to structure experiments and track runs using MLflow
• Techniques for training models using frameworks like scikit-learn, PyTorch, and TensorFlow
• How to manage hyperparameters and perform tuning (including AutoML)
• How to evaluate model performance using key metrics and visualizations
• Best practices for experiment organization, reproducibility, and collaboration
• How trained models are registered and prepared for deployment

This session is designed to help you move from “I have prepared data” to “I can train, evaluate, and track models effectively in Azure ML.” If you’re ready to bring your data to life and start building real machine learning models, this is your next step.