CaseTalk Version 15 Live Demo
At a recent DAMA South Africa session, CaseTalk founder Marco Wobben demonstrated how upcoming version 15 of the information modeling tool bridges the gap between conceptual modeling and Large Language Models — turning static data models into interactive, AI-powered knowledge systems.
The headline feature: CaseTalk now integrates with multiple LLMs (Claude, ChatGPT, Mistral, and others) via the Model Context Protocol (MCP), allowing AI agents to read, query, and reason about your information model in real time.
What does that mean in practice?
An AI That Understands Your Business — Not Just Language
Traditional LLM integrations suffer from hallucinations because the AI lacks context about your specific domain. CaseTalk solves this by exposing its conceptual model through MCP. When you ask the AI "Tell me about Peter," it doesn't guess. It queries the model's API, finds Peter as an example instance, traces his relationships, and responds with verified facts drawn from your own business semantics.
The result: an AI that stays within the boundaries of what the organization has actually defined and confirmed — no hallucinations, no invented data.
A Semantic Layer Without the Semantic Web Complexity
One of the most striking moments in the demo came when host Howard realized the implication: CaseTalk effectively provides a semantic layer — the kind of layer that organizations spend enormous effort building with tools like dbt or PowerBI — but generated directly from the conceptual model.
Even when connected to a messy legacy database like Northwind with cryptic structure, the LLM communicates exclusively in business language. Technical artifacts are hidden behind the conceptual mapping. This is the kind of abstraction that makes AI assistants genuinely useful for business users.
Multiple AI Engines, One Modeling Platform
CaseTalk takes a pragmatic multi-LLM approach:
- Claude excels at logical reasoning, staying on topic, and generating code templates
- ChatGPT shines at creative annotation — generating definitions, examples, and exploratory questions
- Mistral and DeepL handle multilingual translation
Each task within the tool can be configured to use whichever AI performs best for that purpose. Models can be swapped as the AI landscape evolves — no vendor lock-in.
Key Capabilities Demonstrated
- AI Coach: Summarizes your recent modeling work when you open a project, keeping large teams oriented
- Semantic Completion: Suggests alternative natural-language phrasings for fact expressions, helping modelers move beyond generic verbs like "has"
- Example Generation: AI proposes sample data (clearly marked as unconfirmed) to stimulate workshop discussions
- Multilingual Modeling: Translate entire models to other languages with one click — including Afrikaans, Dutch, or even SAP terminology
- Database Reverse Engineering: Drag columns from source systems into your conceptual model, with full lineage tracking
- Artifact Generation: DDL scripts, database views, LinkML, dbt, ArchiMate exports, Data Vault Builder integration, and more — now powered by a templating engine that Claude Code itself helped build
Enterprise-Ready
Version 15 maintains CaseTalk's enterprise foundations: full version control, role-based access, lineage tracking on every element, and difference detection/merge capabilities for team collaboration. Every AI-generated suggestion is clearly marked as unconfirmed until a human approves it.
Why This Matters for Data Management
As DAMA contributors in the audience noted, AI doesn't replace the need for rigorous information modeling — it amplifies it. The modeler's expertise in capturing business semantics becomes the grounding layer that prevents AI from hallucinating. And in return, the AI accelerates the tedious parts: generating definitions, translating terminology, suggesting expressions, and making the model accessible to non-technical stakeholders.
Because even your models deserve a good conversation.
Watch the Full Demo
The full DAMA South Africa session includes live demonstrations of all features described above, plus an interactive Q&A covering topics like semantic collision handling, OWL/Turtle import, multi-user governance, and data obfuscation responsibilities.
Download 
Use Online