AI Hallucination Helps Strengthen Your Information Modelling with Marco Wobben

Key Takeaways

• Integration of LLMs via MCP: CaseTalk uses LLMs via MCP to prevent AI hallucinations by grounding reasoning in factual context.
• Fully Communication-Oriented Information Modelling (FCOIM): The software uses natural-language fact expressions to enable easy verification by subject-matter experts.
• Automated Semantic Layer Creation: The tool integrates a semantic layer into the information model, simplifying traditional Semantic Web methodologies.
• AI as a Collaborative Modelling Assistant: The system uses LLMs such as Claude for logic and ChatGPT for creative tasks and translation.
• Connecting Conceptual Models to Legacy Data: CaseTalk uses FCOIM and MCP to enhance data modelling and effectively prevent AI hallucinations.
• Comprehensive Artefact Generation: The Business resolves semantic conflicts, enhancing clarity and promoting data maturity through agreed-upon definitions.
• Driving Organisational Data Maturity: Semantic conflicts are resolved by the business, ensuring clarity and data maturity in definitions.

Webinar Details

Title: AI Hallucination Helps Strengthen Your Information Modelling with Marco Wobben
Date: 2026-04-13
Presenter: Marco Wobben
Meetup Group: INs and OUTs of Data Modelling
Write-up Author: Howard Diesel

How Does FCOIM Differ from Traditional Ontology Development?

The introductory segment establishes the evolution of the software CaseTalk and its integration with contemporary artificial intelligence. The speaker, Marco Wobben, introduces Fully Communication-Oriented Information Modelling (FCOIM), a methodology originating in the early 1990s that captures precise semantics of business data in natural language.

By implementing the Model Context Protocol (MCP), the system connects AI agents directly to the software’s application programming interface (API). This architecture provides the Large Language Model (LLM) with factual context derived directly from the information model, mitigating the risk of AI hallucinations. Consequently, organisations can generate reliable technical artefacts from a single, verifiable conceptual model.

Figure 1 AI Hallucination Helps Strengthen Your Information Modelling

Figure 2 Brief History

Figure 3 How to Talk to Your Model

Figure 4 FCO-IM is How You Capture What Your Business Actually Means by Its Data

Figure 5 What is FCO-IM?

Figure 6 What is MCP?

Figure 7 The Speaker’s Early Experience of Claude

What is the FCOIM Methodology’s Semantic Layer Approach?

Marco transitions to establishing a semantic layer within the base information model. The FCOIM methodology is contrasted with traditional ontology development using Resource Description Framework (RDF) and Web Ontology Language (OWL). While RDF and OWL typically require technical expertise and operate under an open-world assumption, FCOIM utilises a closed-world approach based on natural language.

Through a practical demonstration involving student residency data, the presentation illustrates how concrete fact statements are translated into comprehensive graph representations and Unified Modelling Language (UML) diagrams. This bottom-up approach ensures that subject matter experts can validate complex models by reading familiar, natural language expressions.

Figure 8 DEMO – in Full Debug Mode

Figure 9 CaseTalk, Lab Edition – Welcome Page

Figure 10 New Expression – City of Residence

Figure 11 Modelling the Expression

Figure 12 Expanding the Model of the Expression

Figure 13 Viewing the Expression as a Fact Model

Figure 14 Viewing the Expression as a UML Class

How Does AI Assist in Managing Complex Models?

The MCP-powered feature analyses the delta of recent model modifications, providing users with a summary of recent sessions to facilitate orientation within extensive models. Furthermore, the presentation addresses interoperability with external semantic frameworks. The software can import Turtle notation files, enabling reverse engineering of semantic web languages. Because semantic web notations often lack natural human semantics, this importation process transitions technical structures into a meta-model, allowing organisations to overlay richer, communicative expressions onto pre-existing data structures.

Figure 15 Model Summary

What are the Advantages of Different LLMs?

While Claude exhibits high proficiency in strict logical reasoning, models like ChatGPT excel in creative augmentation. Applying an LLM to a fundamental entity automatically generates pluralisations, aliases, definitions, and contextually relevant sample questions. These AI-generated prompts are designed to facilitate structured workshops with subject matter experts. Additionally, a semantic completion wizard is demonstrated, wherein the AI proposes alternative phrasings for factual expressions. To maintain structural integrity, any data or examples generated by the AI remain distinctly coded until formally verified by human operators.

Figure 16 Edit Object type ‘City’ – Augment with

Figure 17 Edit Object type ‘City’ – Type & Appearance

Figure 18 Edit population for City

Figure 19 Expression Tree Edit

Figure 20 Semantic Wizard

Figure 21 Semantic Wizard – Verbalisation Suggestions

Figure 22 Verbalisation of Model Apprenticeship

Figure 23 Apprenticeship Model Object Types View

What is Fully Communication-Oriented Information Modelling?

In response to inquiries about semantic ambiguity, Marco demonstrates how MCP integration resolves semantic collisions. Rather than predicting definitions, the AI extracts precise context, proper names, and metadata directly from the defined information model via the API. If overlapping concepts emerge within the same schema, the system alerts users to the inconsistency.

Furthermore, enterprise-level modelling capabilities are discussed, with an emphasis on the need for robust version control in multi-user environments. The software maintains a comprehensive lineage for all modifications, precisely documenting user contributions, AI-generated proposals, and confirmation statuses.

Figure 24 Apprenticeship Model – Chat

How Does CaseTalk Prevent AI Hallucinations Effectively?

The software enables users to map legacy database structures directly onto the conceptual model. This mapping successfully conceals technical artefacts from the LLM; when queried about a specific data point, the AI navigates through conceptual pathways to locate production data without requiring exposure to technical acronyms. Additionally, the presentation highlights robust multilingual support. Utilising AI translation services, entire conceptual models can be translated seamlessly into different languages, instantly updating all corresponding diagrams and expressions.

Figure 25 Data Explorer – Mapping

Figure 26 Data Explorer – City- Data

Figure 27 Fact Table Editor – Translating Language for User Experience

Figure 28 Apprenticeship – Fact Model

Figure 29 Examining Apprenticeship Model

How Does the Software Generate Comprehensive Project Reports?

Expanding upon data capture, the software’s capacity to generate comprehensive project reports is demonstrated. These reports summarise diagrams, terminology, and insights, allowing the LLM to offer targeted, context-aware recommendations for extending the information architecture. Operationally, the tool serves as a bridge to the IT infrastructure, automatically generating Data Definition Language (DDL) schemas, natural-language database views, and data-loading scripts.

A reverse-engineering protocol is also showcased, allowing users to extract columns directly from an external Open Database Connectivity (ODBC) source into the modelling environment. Users can subsequently deploy the AI to refine and correct the narrative structure of the imported relationships.

Figure 30 Apprenticeship Model – Middle Management Overview

Figure 31 Apprenticeship Model – Archimate

Figure 32 Apprenticeship Model – Data Explorer – Multiple Artefacts

Figure 33 Apprenticeship – Customer Model – Fact Model

Figure 34 Customer Model – Mapping

What Role Does AI Play in Data Maturity?

Addressing data security, Marco clarifies that the obfuscation of personal data remains the operational responsibility of the source database and technical connections, rather than the conceptual modelling tool. The software is intended for knowledge capture rather than hosting live production data. Additionally, the rapid evolution of LLMs emphasises the strategic advantage of assigning specific AI models to specialised tasks. To address enterprise security parameters, organisations are encouraged to route API calls through localised LLMs or secure internal cloud environments.

Figure 35 Tool Preferences – Claude

How can Natural Language Improve Semantic Web Technologies?

In the concluding segment, Marco critically evaluates the friction inherent in traditional semantic web technologies. The methodology emphasises that compelling Subject Matter Experts (SMEs) to adopt rigid, technical predicates inhibits effective communication. Instead, modelling should prioritise the natural language that SMEs utilise in daily operations.

Conflicts over organisational terminology should be delegated to business units for resolution rather than assigned to IT departments. Ultimately, integrating AI to augment natural-language modelling provides organisations with a pragmatic foundation to achieve data maturity, ensuring that technical systems accurately reflect operational realities.

Figure 36 “Because even your models deserve a good conversation!”

Figure 37 Casetalk.com – Thank you