OVD-SaaS: Online Verifiable Data Science as a Service, an Architecture of Microservices for Industrial Artificial-Intelligence Applications: Architecture and Study Cases

There is a growing concern about credibility and trustworthiness in results and claims of research in computational data science, and at the same time, difficulty in taking those results to MVP (Minimum viable product) in technology transfer from the laboratory to the industry. Therefore, We present the OVDSaaS prototype, a new platform to manage Data Science Research products (publication, code, data, MVP) that provides reproducibility certification, traceability, provenance, authenticity, legitimacy, reward, FAIR (Findable, Accessible, Interoperable, Reusable) compliance, and valorization to ML/ AI computer scientific research by developing useful online verifiable scientific applications MVP in different industrial domains. To bridge the gap among academia, industry, publishers, and ML/AI technology, we propose an architecture design, methodology, policies, and guidelines for the OVD-SaaS project to advance in reproducibility assessment, persistent identification, validation, and badging of research artifacts results. We present our Scientific Applications case study in the context of reliable scientific research and the validation of results from scientific articles. We describe our experience creating scientific applications and their evolution from simple Demo or non-reproducible artifacts to completely verifiable apps. We analyze common patterns in the creation of scientific applications, discussing the benefits and difficulties encountered that lead to the proposed OVDSaaS features. In this context, four applications were developed for breathing analysis, clinical tumor segmentation, legal assistance, and smart grids, supported in ML/ AI scientific fields such as signal processing, image processing, and NLP. These applications represent end-to-end business solutions cases in the medical, legal, and energy industries. Based on these cases, we performed a reproducibility gap analysis and benchmarking of similar OVDSaaS platforms.