Serbian researchers unveil visual graph editor for edge-cloud workflows

A team of computer scientists from the University of Belgrade has developed a visual graph editor that simplifies the design and deployment of modular applications spanning edge and cloud environments. The editor, presented at the 16th International Conference on Information Society and Technology, addresses the complexity of coordinating multiple computational kernels, shared data, and event-driven execution in distributed systems. Users interact with a unified graphical interface to construct workflows using three first-class abstractions: kernels (computational units), shared memory nodes (distributed data), and event triggers (execution dependencies). Connections define both data flow and control flow, and node properties are configurable within the editor.

The key innovation is automatic serialization: once the graph is complete, the editor outputs machine-readable representations in JSON or XML formats, which can be passed directly to an execution API. This bridges the gap between design-time modeling and runtime deployment, eliminating manual translation steps. The researchers demonstrated the editor's utility with a federated learning workflow, where local training kernels interact with a shared global model via event-driven coordination. Compared to general-purpose diagramming tools, the proposed system provides explicit execution semantics, modularity, and direct deployability. The architecture is extensible, supporting user-defined kernels, event types, and alternative execution backends, making it a flexible foundation for future distributed applications.