How to Develop Robust Semantic Models, Part 1 of 3: From Meaning to Trust
Reading Time: 4 minutes

A modern approach to strengthening meaning and enabling trustworthy AI

Business users want simple answers to apparently simple questions, such as “What counts as a customer, or revenue?” or “Which KPIs can I trust?” and it is up to the data professional, the data architect, or analytics leads, to create the semantic bridge from such questions to the source data, which is often stored across multiple applications, locations, and formats. Semantic integrity is also increasingly important for AI that not only sounds smart but actually understands. Fortunately, the Denodo Platform provides a universal semantic layer that is independent from the physical layer of data source systems, which facilitates this process. 

However, when it comes to developing semantic models, data professionals need a place to begin. In this three-part series, I’ll begin here by discussing a modern Socratic method that can help to enhance trust in the data, for both people and AI. In the second post of this series, I’ll cover how semantic models need to evolve, lest they become rigid and therefore less effective. In Part 3, I’ll conclude with some final thoughts on the critical role of the semantic modeler.  

Semantic Modeling as Modern Maieutics

In classical philosophy, Socratic maieutics is the art of bringing knowledge to light through questioning rather than transmission. Socrates did not position himself as a source of truth, but as a facilitator of understanding, helping others articulate what they implicitly knew. Semantic modeling follows the same principle.

Meaning already exists within organizations, embedded in processes, language, and practices, but it is often tacit, fragmented, and inconsistent. Semantic models do not invent meaning but make it explicit, shared, and operational.

Questioning as a Method for Revealing Intent

A semantic model begins not with data structures, but with questions, like What does a given metric truly represent? Under which assumptions is it valid? Why does a specific event matter in decision-making? For each semantic model, write down as many of these types of questions as you can. Imagine all of the potential personas and use cases, and let these inspire further questions (in a previous post, I talked about the different levels of a semantic model, from the individual applicable data sources to the different ways the data could be used). Don’t be afraid to ask, what other groups would potentially use this data? Next, share your questions, and answers, with a range of domain experts and stakeholders, and incorporate their feedback into new questions and answers. 

This type of disciplined inquiry exposes any hidden ambiguities and contradictions. Through structured dialogue with domain experts, data semantics become explicit and negotiable, and the maieutic process provides a stable reference framework that not only guides further conceptualization but also prevents semantic drift.

Making Tacit Knowledge Explicit

Socrates famously demonstrated that what appears self-evident often dissolves under scrutiny.

In semantic modeling, tacit knowledge plays a similar role, because domain experts rely on experiential understanding that is rarely documented, yet critical for correct interpretation.

A maieutic approach treats this tacit knowledge as the primary input to modeling. By continuously probing definitions and assumptions, the semantic modeler transforms implicit understanding into explicit semantics that can be shared, reused, and governed.

From Dialectic to Conceptual Stability

In maieutics, knowledge advances through dialectic, where propositions are tested, refined, and stabilized through dialogue.

Semantic modeling mirrors this process, if it follows the maieutic process described above. Initial definitions evolve through confrontation with edge cases, conflicting interpretations, and the sometimes unexpected behavior of data in production environments. Coherence and consistency emerge not from rigid upfront design, but from iterative refinement.

The resulting concepts become robust through this  structured questioning and validation process.

Maieutics, Knowledge Engineering, and Semantic Models Design

In modern terms, the maieutic process aligns closely with the principles of knowledge engineering and ontology design, both of which are critical to improve the accuracy and relevance of AI applications. .

Semantic models aim to formally represent shared conceptualizations of a domain, but their quality depends on how well implicit knowledge is elicited and validated. Maieutic questioning provides the foundation for this process, because competency questions, concept hierarchies, constraints, and relationships are not merely technical artifacts, but are the formal outcomes of the guided discovery of meaning, based on iterative questioning.

In this sense, semantic modeling acts as applied epistemology, translating human understanding into machine-interpretable structures, without losing conceptual fidelity.

Formalizing Meaning Without Freezing It

One of the central challenges in knowledge engineering is balancing formalization and adaptability.

A maieutic approach helps avoid premature rigidity by enabling formal definitions to remain traceable to the original intent, and logical constructs, mappings, and constraints become expressions of meaning rather than static abstractions.

This preserves the semantic model’s ability to evolve as organizational knowledge and context change, without sacrificing coherence.

Governance as Ethical Semantics

Socratic philosophy links knowledge with responsibility. Similarly, semantic governance is not merely a technical safeguard, but an ethical commitment to correct misinterpretation.

Access controls, security rules, data masking, quality constraints, and lineage are ways of enforcing meaning in operational contexts. Data governance enables semantics to be applied consistently and transparently, preventing misuse and further misinterpretation.

In this sense, data governance is the operational continuation of the maieutic process.

The Semantic Modeler as a Contemporary Socratic Figure

This perspective redefines the role of the semantic modeler.

Rather than acting as a passive translator of requirements, the modeler becomes a facilitator of understanding, a challenger of assumptions, and a guardian of meaning. Working alongside domain experts, data engineers, and governance roles, the semantic modeler helps the organization articulate what it truly means by its data, transforming individual perspectives into shared knowledge.

Enabling Explainable and Trustworthy AI

Socrates warned against false knowledge that imitates wisdom without substance.

In the age of AI, this warning is especially relevant, because AI systems built on poorly articulated semantics risk producing outputs that appear plausible but lack grounding.

A maieutic semantic layer provides explicit, validated meaning, reducing ambiguity, bias, and hallucinations. For generative AI and RAG applications, this semantic grounding is essential for explainability and trust.

Maieutics in Action

Denodo operationalizes this modern maieutic approach through its universal semantic layer.

By separating meaning from physical data, Denodo provides the space where intent can be clarified, concepts stabilized, and data governance embedded. The semantic layer becomes the locus where tacit knowledge is made explicit, aligned, and enforced.

For analytics, AI, and generative AI use cases, this means that the AI can consume data that is already contextualized, governed, and semantically coherent.

In a landscape where AI effectiveness depends increasingly on the quality of meaning rather than the quantity of data, Denodo’s approach positions the semantic layer as a critical enabler of trustworthy, explainable, and scalable intelligence. In my next post in this series, I’ll cover the evolutionary nature of semantic modeling.

Andrea Zinno
Latest posts by Andrea Zinno (see all)