Tutorials

The shift toward distributed, latency-sensitive applications is accelerating the adoption of serverless computing across the edge-cloud continuum. Yet most serverless platforms remain energy-agnostic, optimising for cost and responsiveness while overlooking the energy implications of fine-grained function placement. This tutorial introduces the principles, architectures, and emerging techniques behind energy-aware serverless orchestration, focusing on how to schedule and execute functions across heterogeneous edge and cloud environments under dynamic energy, and resource constraints. We cover energy-adaptive placement strategies, energy-aware scheduling, lightweight serverless runtimes, and telemetry-driven optimisation using real-time power and performance signals. Through conceptual walkthroughs and a practical case study on Software Defined Networks, participants will learn how to design sustainable, efficient, and policy-compliant serverless pipelines for modern distributed workloads. The tutorial concludes with an open research agenda spanning energy modelling and cross-layer orchestration in energy-aware edge-cloud systems.

Doing research with AI agents is often framed as a matter of combining effective prompts with specialized skills. We argue that this overlooks a more fundamental component: an explicit blueprint of how a discipline produces and validates knowledge. This tutorial presents such a blueprint as an inspectable and reproducible graph of research steps, decision gates, and iterative loops. Its building blocks include literature retrieval with explicit acceptance criteria; graph-based memory structured around a paper ontology; hypothesis loops in which proposing agents are challenged by independent refuting agents equipped with external verifiers; context partitioning to control information flow; and observability mechanisms for inspecting intermediate decisions and artifacts. The central premise is that agentic research workflows should be designed not only to produce scientific results, but also to make the process that generates them measurable, testable, and reproducible. The blueprint is therefore not merely scaffolding around the science, but a scientific artifact itself.