Operational Data-Centric AI: A Review of Mesh Governance, Synthetic Tables, Efficient Transformers, Distribution Shift, and Decision Automation
Main article
Abstract
The 2024 DATAMIND corpus moves from foundational database-centered artificial intelligence toward operational data-centric AI. This review analyzes all DATAMIND articles published in 2024 and situates them within eighty DOI-bearing references on data management, synthetic tabular data, efficient transformer architectures, digital twins, concept drift, uncertainty, portfolio optimization, and reinforcement learning. The review uses a structured coding matrix to compare each article by lifecycle stage, architectural object, risk category, decision setting, and evaluation emphasis. The analysis shows that DATAMIND's 2024 articles collectively frame operational AI as a control system composed of data products, generative data, scalable models, monitoring routines, and decision rules. DataMesh research emphasizes distributed ownership and accountability; synthetic tabular data research highlights privacy and utility trade-offs; efficient transformer work reframes computation as a governance resource; smart manufacturing connects AI to cyber-physical evidence streams; distribution shift makes monitoring central; and actor-critic portfolio optimization brings decision value into focus. Two high-resolution grayscale figures and three tables summarize the journal corpus, coding rubric, comparative strengths, and research agenda. The article concludes that data-centric AI should be assessed by reproducibility, scalability, robustness, governance, and decision value rather than by model accuracy alone.
