Engineering Trust in AI Systems: A Data-Layer Framework for Explainability and Auditability
Abstract
The article examines engineering approaches for strengthening trust in AI systems through data-layer controls that make decisions explainable and verifiable in audits. The widening regulatory and organizational demands for the traceability of training data, the reproducibility of pipelines, and the defensible documentation of model behavior in production drive practical relevance. Scientific novelty lies in integrating provenance capture, lineage graphs, feature-store governance, and standardized documentation artifacts into one coherent data-layer framework that produces machine-checkable evidence. The work describes lifecycle evidence generation from ingestion to inference, studies how provenance models and lineage datasets support inspection, and analyzes how documentation instruments complement technical traces. Special attention is given to preventing evidence gaps caused by opaque preprocessing, weak versioning, and incomplete logging. The study aims to systematize a data-layer architecture that supports explainability and auditability without relying on new model classes. A comparative analysis of recent research, a synthesis of published frameworks, and a structured review of sources are employed in this study. The conclusion summarizes actionable controls and their expected audit outputs. The article targets engineers, MLOps teams, risk functions, and internal/external auditors.
Keywords
Downloads
References
- Ahmed, M., Dar, A. R., Helfert, M., Khan, A., & Kim, J. (2023). Data provenance in healthcare: Approaches, challenges, and future directions. Sensors, 23(14), 6495. https://doi.org/10.3390/s23146495
- Chen, Y., Zhao, Y., Li, X., Zhang, J., Long, J., & Zhou, F. (2024). An open dataset of data lineage graphs for data governance research. Visual Informatics, 8(1), 1–5. https://doi.org/10.1016/j.visinf.2024.01.001
- de la Rúa Martínez, J., Buso, F., Kouzoupis, A., Ormenisan, A. A., Niazi, S., Bzhalava, D., Mak, K., Jouffrey, V., Ronström, M., Cunningham, R., Zangis, R., Mukhedkar, D., Khazanchi, A., Vlassov, V., & Dowling, J. (2024). The Hopsworks feature store for machine learning. In Companion of the 2024 International Conference on Management of Data (SIGMOD ’24) (pp. 135–147). Association for Computing Machinery. https://doi.org/10.1145/3626246.3653389
- Gilbert, S., Adler, R., Holoyad, T., & Weicken, E. (2025). Could transparent model cards with layered accessible information drive trust and safety in health AI? npj Digital Medicine, 8(1), 124. https://doi.org/10.1038/s41746-025-01482-9
- Kalokyri, V., Tachos, N. S., Kalantzopoulos, C. N., Sfakianakis, S., Kondylakis, H., Zaridis, D. I., Colantonio, S., Regge, D., Papanikolaou, N., Marias, K., Fotiadis, D. I., Tsiknakis, M., & (2025). AI model passport: Data and system traceability framework for transparent AI in health. Computational and Structural Biotechnology Journal, 28, 386–404. https://doi.org/10.1016/j.csbj.2025.09.041
- Liu, R., Park, K., Psallidas, F., Zhu, X., Mo, J., Sen, R., Interlandi, M., Karanasos, K., Tian, Y., & Camacho-Rodríguez, J. (2023). Optimizing data pipelines for machine learning in feature stores. Proceedings of the VLDB Endowment, 16(13), 4230–4239. https://doi.org/10.14778/3625054.3625060
- Longpre, S., Mahari, R., Chen, A., et al. (2024). A large-scale audit of dataset licensing and attribution in AI. Nature Machine Intelligence, 6, 975–987. https://doi.org/10.1038/s42256-024-00878-8
- Mökander, J., Schuett, J., Kirk, H. R., et al. (2024). Auditing large language models: A three-layered approach. AI Ethics, 4, 1085–1115. https://doi.org/10.1007/s43681-023-00289-2
- Schlegel, M., & Sattler, K.-U. (2025). Capturing end-to-end provenance for machine learning pipelines. Information Systems, 132, 102495. https://doi.org/10.1016/j.is.2024.102495
- Staufer, L., Yang, M., Reuel, A., & Casper, S. (2025). Audit cards: Contextualizing AI evaluations (arXiv:2504.13839). arXiv. https://arxiv.org/abs/2504.13839