Orchestrating Safe & Adaptive Learning with Multi-Agent Systems
Details
Join the Orchestrating Safe & Adaptive Learning with Multi-Agent Systems by Josephat Onkoba on August 29, 2026, at 6 PM CET Malta time.
As AI agents transition from novelty to necessity in higher education, developers face a critical "sociotechnical gap": how to build autonomous systems that personalize learning without hallucinating or violating institutional policies. In this session, we explore a design-oriented framework for integrating Multi-Agent Systems (MAS) into Learning Management Systems (LMS) using the von Neumann Multi-Agent Framework (vNMF).
Moving beyond simple chatbots, we will deconstruct a sophisticated architecture where specialized agents, including a "Pedagogical Agent" for scaffolding and a "Guardian Agent" for governance, collaborate to optimize student outcomes.
What you will learn:
- The vNMF Blueprint: How to architect agents by decoupling Control, Logic, Memory, and I/O units to mirror computer architecture for better state management and reliability.
- Orchestration Patterns: Practical implementation of the "Coordinator/Dispatcher" and "Generator-Critic" patterns using frameworks like LangGraph to manage complex workflows and reduce latency.
- Solving the "Overreliance" Problem: Strategies for implementing "cognitive forcing functions" and "appropriate reliance" checks to prevent students from blindly accepting AI hallucinations.
- Policy-as-Code: How to deploy a "Guardian Agent" that technically enforces data privacy and academic integrity rules in real-time, moving compliance from a policy document to the codebase.