Course Code: advautogen
Duration: 14 hours
Prerequisites:
- Proficiency in Python programming
- Experience building with LLM-based applications
- Familiarity with function calling and multi-agent system design
Audience
- Senior developers
- Platform engineers
- AI architects
Overview:
AutoGen is an open-source framework from Microsoft for building multi-agent applications that use LLMs, tools, memory, and user interaction.
This instructor-led, live training (online or onsite) is aimed at advanced-level developers and architects who wish to design and deploy deeply customized agents using AutoGen’s Python-based APIs, function-calling capabilities, and modular toolchains.
By the end of this training, participants will be able to:
- Develop custom agents with role-specific logic and tool routing.
- Build dynamic workflows using advanced function calling and context switching.
- Implement memory modules and planning frameworks within agent teams.
- Handle multi-agent error states and adaptive retry mechanisms.
Format of the Course
- Interactive lecture and discussion.
- Lots of exercises and practice.
- Hands-on implementation in a live-lab environment.
Course Customization Options
- To request a customized training for this course, please contact us to arrange.
Course Outline:
Review of AutoGen Core Concepts
- Agent and group definitions
- Function calling and role chaining
- Limitations of built-in agents and where customization is needed
Building Custom Agents with Python
- Defining agent behavior using user_proxy and AssistantAgent subclasses
- Injecting role-specific logic and decision-making
- Creating reusable agent modules and mixins
Advanced Tool Integration and Routing
- Tool registration, binding, and invocation
- Conditionally routing inputs to specific tools
- Managing multi-step toolchains and composite actions
Planning and Context Management
- Designing task decomposers and intermediate planners
- Maintaining context across chained agents
- Implementing scoped memory for long-running sessions
Error Handling and Recovery Mechanisms
- Detecting and managing failed or incomplete interactions
- Agent-triggered retries and fallback logic
- Logging, debugging, and response validation
Multi-Agent Collaboration with Custom Roles
- Coordinating specialists within dynamic agent groups
- Orchestrating reasoning loops and cooperative workflows
- Role separation vs. role blending in task assignments
Real-World Deployment Strategies
- Optimizing for performance and cost (token use, caching)
- Embedding AutoGen workflows into web apps or pipelines
- Security, observability, and user feedback integration
Summary and Next Steps