Back to homepage
Blog March 24, 2025 7 min read

Knowledge Graph Instead of Data Chaos: How AI Turns Meetings into Real Knowledge

8 months, every meeting transcribed, automatically enriched, and fed into a knowledge graph. The result: answers based on real knowledge – not guesswork.

Roberto Schlatter
Roberto Schlatter Founder & AI Consultant, Moro Vision GmbH

I have been transcribing every single meeting for the past 8 months. Not just the important ones. Not 50 percent. All of them.

Each transcription is automatically enriched with metadata, summarized, and fed into a vector database with an attached knowledge graph. My AI platform is connected to this graph.

Meetings are just the starting point – a particularly effective entry point because most implicit knowledge emerges in conversations. However, the same approach can be applied to virtually any knowledge source: contracts and proposals, internal documentation and wikis, email correspondence, support tickets, project reports, or regulatory documents such as SOPs and audit protocols. Wherever knowledge is fragmented across silos today, a knowledge graph can reveal connections that would otherwise be lost.

The Result: Answers in Seconds

I ask a question and get answers drawn from 8 months of conversations. Precise. With context and nuance. Without hallucinations.

MV-insights
What does the final agenda for the AI customer event look like? Here is the final agenda in detail:
AI Customer Event – Agenda
Wednesday, March 4 approx. 30 participants
TimeAgenda Item
08:00Welcome & Opening
MorningFundamentals & Case Studies – Machine Learning, Deep Learning, Generative AI, Prompt Engineering + real-world case studies
12:00–13:00Lunch Break
AfternoonInteractive Workshop Stations (max. 3–4 participants per station)
AfterwardsNetworking
Source: 3 meetings (Jan–Mar) · Last updated Mar 12
What IP concerns did the client raise? The following concerns were raised across 3 separate meetings:
  1. Training data storage location – Where is the data physically stored? Requirement: Swiss data center.
  2. Usage rights for fine-tuning – Who owns the trained models and weights?
  3. Data portability upon contract termination – How are data and models fully transferred in an exit scenario?
Source: Meetings Jan 14, Feb 28, Mar 6 · Priority: High
What needs to be ready by March 2? 4 open tasks identified:
TaskStatus
Final slide deckDone
Attendee list incl. dietary restrictionsDone
Budget approval for cateringOpen
Floor plan with AV setupOpen
Source: Last team meeting · Status: 2/4 completed

Answers in seconds. Based on real knowledge. Not guesswork.

What Is a Knowledge Graph – and Why Does It Matter?

The core of this system is a knowledge graph. Not a buzzword – a data structure.

Nodes

Real-world entities: people, companies, topics, decisions, projects.

Edges

Relationships between entities: "works at", "decided", "contradicts".

Ontology

A set of rules that defines how these entities relate to each other and which relationship types are valid.

The difference from a conventional database: A knowledge graph does not just understand that two things are connected. It understands how and why. This makes it the ideal foundation for AI systems that do not merely search, but reason.

From fragmented knowledge to a strategic network: Pipeline overview with data silos, ontology, knowledge graph, and AI integration
Fig. 1 — From fragmented knowledge to a strategic network: Data silos are structured through an ontology, linked in a knowledge graph, and made accessible to AI systems.

The Problem: Mountains of Unstructured Knowledge

Most organizations sit on mountains of unstructured knowledge. Meetings, emails, documents, Slack messages. Everything is there. Nothing is connected. No system that creates meaning.

Conventional Approach Knowledge Graph Approach
Data in silos (email, docs, chat) All sources linked in a single graph
Keyword-based search Semantic queries based on meaning
Context is lost Relationships are preserved
AI hallucinates when data is missing AI reasons from real knowledge
Information becomes outdated unnoticed Temporal dimension is built in

Raw model intelligence will not solve this. GPT-5, Opus 4, or Gemini 3 will not solve it either. Because the model is not the differentiator – your data and how it is connected are.

Visualization of a real knowledge graph from 8 months of meeting transcriptions with interconnected entities
Fig. 2 — A real knowledge graph in action: Hundreds of interconnected entities from 8 months of meeting transcriptions, filtered by my name. Each node represents a person, a company, a topic, or a decision.

What Makes the Difference

What truly moves organizations forward are not bigger models. It comes down to three things:

Structure

Data needs a schema, an order. Without structure, everything remains fragmented – no matter how intelligent the model is.

Context

Who said what, and when? In what context? Context turns data into information and information into knowledge.

Relationships

The connections between pieces of information are often more valuable than the information itself. Relationships enable reasoning.

In a world where intelligence is becoming a commodity, meaning is the scarcest resource. Those who structure it, win.

Closing the Loop: From Graph to Language Model

However, that is only half the journey. Transcription, enrichment, and storage in the knowledge graph form the foundation – but to close the loop, this structured knowledge must be made accessible to the language model of choice. As an official partner of Langdock, we connect GDPR-compliant, EU-hosted models via MV-insights directly to the knowledge graph. This closes the circle: employees can effectively and precisely query their corporate knowledge – and receive concrete, accurate answers without hallucinations.

The Technical Architecture Behind It

For the implementation, I use a privacy-compliant infrastructure that runs entirely on Swiss and European servers:

  • Transcription & Summarization: Automated real-time transcription and summarization of meetings using the GDPR-compliant tool Sally.io
  • Metadata Enrichment & Automation: Workflow automation with n8n
  • Vector Database: Semantic search across all transcribed content
  • Knowledge Graph: Linking all entities with typed relationships
  • AI Platform: GDPR-compliant language model with direct access to graph context via Langdock

The entire setup is compliant with the Swiss Data Protection Act (DSG) and can also be operated on-premise – a decisive factor for regulated industries.

Curious what this looks like in practice?

Find out how a knowledge graph system can be implemented in a privacy-compliant way for your organization.

Book a free consultation → Learn more about Moro Vision →