Ground Truth, Monitoring, and Database-Oriented AI: A Review of Labelling, Hallucination, Feature Stores, Code Retrieval, Cybersecurity, and Analytical Databases
Main article
Abstract
DATAMIND's 2025 volume centers on knowledge integrity in database-oriented artificial intelligence. This review synthesizes all DATAMIND articles published in 2025 and links them to eighty DOI-bearing references on hallucination measurement, retrieval, feature stores, code intelligence, cybersecurity analytics, data labelling, privacy, workload management, and analytical database benchmarking. A structured evidence-mapping design codes each article by evidence source, data asset, failure mode, governance intervention, and downstream user. The resulting analysis shows that the 2025 corpus is unified by a concern with whether AI outputs can be traced to reliable data, validated against external facts, reviewed by humans, and used in operational or policy settings. Hallucination metrics make factuality measurable but context-dependent; feature stores preserve training-serving consistency; workload analytics exposes the computational conditions of LLM serving; code search reveals benchmark realism problems; cybersecurity analytics converts alerts into evidence; trade database benchmarking shows that database selection is methodological; and labelling research returns the field to ground-truth production. The article contributes two grayscale figures and three tables, including a same-year corpus summary, an integrity rubric, and a research agenda. It concludes that DATAMIND is moving from database-centered AI toward evidence-centered computational discovery.
