A DataOps Toolbox Enabling Continuous Semantic Integration of Devices for Edge-Cloud AI Applications
Mario Scrocca, Marco Grassi, Alessio Carenini, Jean-Paul Calbimonte, Darko Anicic, Irene Celino
TL;DR
Heterogeneous devices and data formats across edge-to-cloud AI deployments impede rapid development and interoperability. The paper proposes Continuous Semantic Integration (CSI), a Semantic Web–driven approach coupled with a modular DataOps toolbox to enable low-code, semantically interoperable data exchanges and knowledge-graph–based representations. It demonstrates three cross-domain use cases (manufacturing, mobility, rehabilitation) with pipelines that harmonize static descriptions and runtime data using OPC UA, WoT, SOSA, and ASAM vocabularies, supported by declarative mappings and reusable templates. The work shows practical viability and stakeholder uptake, highlighting benefits for scalability, configurability, and potential integration with Large Language Models for natural-language-driven queries and pipeline customization.
Abstract
The implementation of AI-based applications in complex environments often requires the collaboration of several devices spanning from edge to cloud. Identifying the required devices and configuring them to collaborate is a challenge relevant to different scenarios, like industrial shopfloors, road infrastructures, and healthcare therapies. We discuss the design and implementation of a DataOps toolbox leveraging Semantic Web technologies and a low-code mechanism to address heterogeneous data interoperability requirements in the development of such applications. The toolbox supports a continuous semantic integration approach to tackle various types of devices, data formats, and semantics, as well as different communication interfaces. The paper presents the application of the toolbox to three use cases from different domains, the DataOps pipelines implemented, and how they guarantee interoperability of static nodes' information and runtime data exchanges. Finally, we discuss the results from the piloting activities in the use cases and the lessons learned.