PANDORA D2.2: Architecture and Technical Specifications

PANDORA aims to develop a comprehensive AI-based and domain-informed framework, with the objectives of improving the preparation and the delivery of complete and trustworthy datasets for training and enhancing AI models deployed in AIoT systems, and increasing the degree of autonomy, trustworthiness, and efficiency for deploying and operating these AIoT systems. We address the first objective in Phase 1 of PANDORA with the following strategies: synthetic data generation, semi-annotations of datasets, uncertainty quantification, data summarization, and verification and validation. The second objective is addressed in Phase 2 which comprises AI as a Service (AIaaS) and Computing as a Service (CaaS) components to support continual, energyefficient, and robust AIoT operation. Thereport analyzes each of the components involved in the project, detailing their functionalities, technical specifications, and dependencies. We rely on the 4+1 Architectural View Model to present the PANDORA architecture from different perspectives, including logical, development, process, data, and physical views. The report includes an integration plan that outlines the software engineering approach, including agile and waterfall methodologies, for developing and deploying the PANDORA framework. Finally, we conclude with the technicalrequirements for each PANDORA component.