Orchestration patterns with the Agent Framework – Multi-agent systems (AI-500, lesson 15)

What you learn in this lesson

map the five patterns of the exam (hub-and-spoke, sequential, parallel, peer-to-peer, orchestrator-subagent) to the five orchestrations of the Agent Framework and say where Learn names no equivalent

build the flow of the group-travel team from lesson 1.1 in stages with Sequential and Concurrent

set the revision loop with a limit in code and check the call counter against it

set up Handoff, Group Chat and Magentic and justify why the case does not use them

write a decision note on the choice of patterns and try it on a new case

Lesson 1.1 fixed the flow of the group-travel team of Nordlicht Reisen GmbH: rule in code, planer, then unterkunft and transport in parallel, then kalkulation, recht, texter and pruefer one after another, at most one revision, then the trip planner (M-02, F-01). So far that is a table. This lesson builds the flow with the Microsoft Agent Framework and answers a question of the exam: What are the patterns you have to know for it, and what does the framework bring? The exam uses other names than the framework. The lesson maps the two lists to each other and says open where Learn names no mapping.

Prior knowledge. "AI track, lesson T7.6" introduces the five orchestrations of the framework (sequential, concurrent, handoff, group chat, Magentic), works out that cost, errors and waiting time grow with every agent, and builds a checker. "T7.4" shows workflows made of executors and edges, "AI-103, lesson 2.5" the control in code and the checker layer. Lesson 1.1 designed the flow and lesson 0.2 the LaufZaehler. What goes beyond that starts here: the names of the exam, several patterns in one flow and the limits of the loop in code.

New in this lesson. The flow is the class AngebotsFluss. The case stays as in 1.1; new is the implementation in stages: a stage is a workflow of the framework, and the code between the stages decides. The revision runs at most once (MaxUeberarbeitungen = 1). In the run with fake clients (one call per agent) there are 7 calls, with a revision 9, under the upper limit of 20 (lesson 0.2).

The requirement

All numbers are assumptions of the task.

No.	Requirement	Type	Measure	Weight	Source
2.8-01	The order of the steps is in code, not in the model.	R	0 steps chosen by a model	Must	F-01
2.8-02	`unterkunft` and `transport` run in parallel.	Q	2 agents in one stage	Should	M-02 (under 10 minutes)
2.8-03	The revision between `texter` and `pruefer` is limited.	R	at most 1 revision, then a person	Must	1.1-06
2.8-04	A run makes at most 20 model calls.	R	20 (lesson 0.2)	Must	M-11
2.8-05	Legal questions and refunds reach no model.	R	0 calls	Must	M-03, M-05

This lesson belongs to course AI-500

You will see the rest of this lesson and its video once you have bought the course. The goals above show what you learn here.

€129.00

incl. VAT, one-time payment

Access to all of the course's lessons with their exercises, solutions and videos on this site, for 24 months from the purchase. A self-study course: no tutoring, no exams or grading by us, and no certificate.

Withdraw from contract here