AI-Ready Data Products: Extending Data Contracts and Semantic Layers for LLM and Agent Consumption
DOI:
https://doi.org/10.15680/IJCTECE.2025.0804024Keywords:
data products, data contracts, semantic layer, LLM agents, Model Context Protocol, text-to-SQL, data governanceAbstract
Enterprise data products have been designed for human analysts and business-intelligence tools, which bring tacit knowledge to every query. Large language model (LLM) agents that plan, query and act on enterprise data do not have that knowledge, and they expose gaps in current data contracts: metric definitions that live only in dashboards, undeclared grain, undocumented business rules, ambiguous business vocabulary, and access permissions that do not carry through into agent tool calls. This paper proposes an AI-ready data product specification that extends conventional data contracts with machine-interpretable entity and grain declarations, join paths with cardinality, governed metric definitions, synonym bindings, value domains, runtime quality and freshness signals, usage constraints and an agent-facing interface exposed through the Model Context Protocol (MCP). We map the specification onto a lakehouse-plus-semantic-layer architecture with a governance plane built around a catalog such as Microsoft Purview, and we define controls for auditing agent access and for preventing privilege escalation through tool chaining. The design is illustrated with order-to-cash, supply-chain and customer-experience data products from a global industrial manufacturer. A representative evaluation with a GPT-4o ReAct agent answering 120 business questions over four data products (360 runs per condition) compares raw schema, schema plus a conventional contract, and the full specification. Execution accuracy rises from 41.9% to 53.9% to 79.7%, with the largest gains on metric-with-business-rule and ambiguous-term questions, while semantically wrong but executable answers fall from 41.1% to 11.9% of runs and policy-violating tool calls fall from 31 to 7
References
[1] Z. Dehghani, Data Mesh: Delivering Data-Driven Value at Scale. Sebastopol, CA, USA: O'Reilly Media, 2022.
[2] M. Armbrust, A. Ghodsi, R. Xin, and M. Zaharia, "Lakehouse: A new generation of open platforms that unify data warehousing and advanced analytics," in Proc. Conf. Innovative Data Systems Research (CIDR), 2021.
[3] S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y. Cao, "ReAct: Synergizing reasoning and acting in language models," in Proc. Int. Conf. Learning Representations (ICLR), 2023.
[4] L. Wang et al., "A survey on large language model based autonomous agents," arXiv preprint arXiv:2308.11432, 2023.
[5] Anthropic, "Introducing the Model Context Protocol," announcement and specification v2024-11-05, Nov. 2024.
[6] A. Floratou et al., "NL2SQL is a solved problem... Not!," in Proc. Conf. Innovative Data Systems Research (CIDR), 2024.
[7] J. Sequeda, D. Allemang, and B. Jacob, "A benchmark to understand the role of knowledge graphs on large language model's accuracy for question answering on enterprise SQL databases," arXiv preprint arXiv:2311.07509, 2023.
[8] A. Jones, Driving Data Quality with Data Contracts. Birmingham, U.K.: Packt Publishing, 2023.
[9] Bitol Project (LF AI & Data Foundation), "Open Data Contract Standard (ODCS)," version 2.2.1, Dec. 2023. [Online]. Available: https://github.com/bitol-io/open-data-contract-standard
[10] PayPal, "Data contract template," GitHub repository, 2023. [Online]. Available: https://github.com/paypal/data-contract-template
[11] R. Kimball and M. Ross, The Data Warehouse Toolkit: The Definitive Guide to Dimensional Modeling, 3rd ed. Indianapolis, IN, USA: Wiley, 2013.
[12] dbt Labs, "About MetricFlow" and "dbt Semantic Layer," product documentation, 2023.
[13] Snowflake Inc., "Cortex Analyst: Semantic model specification," product documentation, 2024.
[14] T. Yu et al., "Spider: A large-scale human-labeled dataset for complex and cross-domain semantic parsing and text-to-SQL task," in Proc. Conf. Empirical Methods in Natural Language Processing (EMNLP), 2018, pp. 3911–3921.
[15] J. Li et al., "Can LLM already serve as a database interface? A BIg bench for large-scale database grounded text-to-SQLs," in Proc. Advances in Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track, 2023.
[16] M. Pourreza and D. Rafiei, "DIN-SQL: Decomposed in-context learning of text-to-SQL with self-correction," in Proc. Advances in Neural Information Processing Systems (NeurIPS), 2023.
[17] D. Gao, H. Wang, Y. Li, X. Sun, Y. Qian, B. Ding, and J. Zhou, "Text-to-SQL empowered by large language models: A benchmark evaluation," Proc. VLDB Endowment, vol. 17, no. 5, pp. 1132–1145, 2024.
[18] T. Schick, J. Dwivedi-Yu, R. Dessì, R. Raileanu, M. Lomeli, L. Zettlemoyer, N. Cancedda, and T. Scialom, "Toolformer: Language models can teach themselves to use tools," in Proc. Advances in Neural Information Processing Systems (NeurIPS), 2023.
[19] P. Lewis et al., "Retrieval-augmented generation for knowledge-intensive NLP tasks," in Proc. Advances in Neural Information Processing Systems (NeurIPS), 2020, pp. 9459–9474.
[20] K. Greshake, S. Abdelnabi, S. Mishra, C. Endres, T. Holz, and M. Fritz, "Not what you've signed up for: Compromising real-world LLM-integrated applications with indirect prompt injection," in Proc. 16th ACM Workshop on Artificial Intelligence and Security (AISec), 2023, pp. 79–90.
[21] OWASP Foundation, "OWASP Top 10 for Large Language Model Applications," version 1.1, 2023.
[22] National Institute of Standards and Technology, "Artificial Intelligence Risk Management Framework (AI RMF 1.0)," NIST AI 100-1, Jan. 2023.
[23] M. Jones, A. Nadalin, B. Campbell, J. Bradley, and C. Mortimore, "OAuth 2.0 Token Exchange," IETF RFC 8693, Jan. 2020.
[24] Microsoft, "Microsoft Purview data governance documentation," product documentation, 2024.
[25] OpenAI, "GPT-4o system card," arXiv preprint arXiv:2410.21276, 2024.
