Event Details

Modern distributed applications increasingly combine services across cloud, edge, IoT, and 5G/6G infrastructures. Automating such workflows requires coordinating heterogeneous service models, provider constraints, management interfaces, and interaction mechanisms across multiple domains. Yet workflow automation and service composition remain largely fragmented, manual, and expert-driven, particularly when users express their needs in natural language. Such intents may be incomplete, ambiguous, or formulated at different levels of abstraction, while large language models may generate nonexistent services, invalid configurations, or unsupported assumptions.

This talk presents KASCADE—Knowledge-grounded Agentic Service Composition Across Distributed Environments—a vision for moving beyond conventional workflow automation toward intent-driven, knowledge-grounded, and adaptive workflow automation. KASCADE views service composition as an end-to-end lifecycle that connects intent interpretation, workflow and composition plan construction, service discovery and selection, intent negotiation and reporting, deployment, execution, and runtime intent assurance. Its architecture combines specialized agents with explicit knowledge about service capabilities, provider offerings, constraints, and management operations. This separation enables the system to distinguish what users want from how their needs can be realized, while adapting interpretation and composition to the abstraction level of each request.

Within this framework, LLM-based agents provide flexible natural-language and task-level reasoning, while structured knowledge grounds their decisions, reduces hallucinations, and guides the progressive construction of valid and executable workflows. Multi-agent coordination further enables complex composition tasks to be decomposed into specialized reasoning activities that can be autonomously executed and dynamically coordinated across heterogeneous domains.

The talk will conclude with a demonstration of the current KASCADE proof of concept, starting from a natural-language intent and automatically deploying a 5G service composed of a 5G Core sub-slice and a RAN sub-slice. The demonstration illustrates how intent-driven workflow automation can progressively transform high-level user requirements into concrete, executable service compositions, providing a step toward more automated, adaptive, traceable, and robust workflow management across distributed environments.

Zoom Link:

Speaker Bio

Walid's CV, and webpage: https://www-public.imtbs-tsp.eu/~gaaloulw/