01
Three search principles, different strengths
Compare BM25 for term matching, vectors for semantic similarity and graphs for modeled relationships.
Keynote / Search & Retrieval
Enterprise Search: Not every question needs a vector.
BM25, vector search and graphs answer different kinds of questions. Using accessible examples, the talk lays out their strengths and limits, and shows when a combination makes more sense than searching for the one best method.
45–60 minGerman / EnglishFee from €6,500 plus VAT and travel
Who it’s for
Who this keynote is for.
For IT and data leaders, architects, and teams planning enterprise search or knowledge-based AI applications.
What your audience takes away.
- A clear understanding of the differences between lexical, semantic and graph-based search.
- Decision criteria grounded in user questions, data and operational requirements.
- An approach for comparing search quality with relevant examples instead of gut feeling.
Content framework
Three perspectives on the topic.
We tailor the focus and technical depth to your question.
02
When to use which, and when to combine?
Discuss with examples how the question and the data shape the choice of a method or a combination.
03
Search quality can be tested
Treat relevance, source quality, access rights and freshness as criteria for a robust search solution.
How I work
From search query to the right context.
This simplified example shows how sources and follow-up questions narrow down an answer. It is not a search benchmark and not a performance comparison of BM25, vector and graph search. We assess the right approach based on your questions and data.
Retrieval in action
Explore the workflow with curated, fictional sample data, fully local and without a live LLM.
- 1Classify
- 2Retrieve sources
- 3Combine
- 4Answer
The example answer is ready.
01 / Classification
Knowledge question · retrieval architecture · conceptual
02 / Retrieve sources in parallel
Vector search
Matching text passagesExample text on semantic search and context-based answers.
Knowledge graph
Entities & relationshipsDocument → product → version → responsible team.
Problem-solution graph
Linked solution pathsNot needed for this question
Structured data
Versions & statusNot needed for this question
03 / Synthesis
Combine relevant results and make the limits of the answer visible.
04 / Answer with sources
Vector search can find passages with similar content. A knowledge graph adds explicit relationships, for example between document, product and version. That can help narrow down the right context. The benefit depends on modeling and data quality; a graph does not guarantee a correct answer.
- Vector search
- Knowledge graph
What would this look like with your documents? In the Machine-Readability Check, we examine your knowledge foundation and the next steps.
Request a workshop proposal →Before you inquire
Frequently asked questions.
How long is a keynote?
The keynote format runs 45–60 minutes. Alternatives within the offering are a 20-minute talk, a 90-minute deep dive with Q&A, or a panel by arrangement.
What does the keynote cost?
Fee from €6,500 plus VAT and travel. International €8,500–12,000 plus VAT. All fees plus VAT.
In which languages can you give the talk?
I speak fluent English. All offerings and intro calls are available in German and English.
Can the keynote be delivered remotely?
Yes. Keynotes are available on-site and remotely; just mention your preferred setting in your inquiry.
How technical is the talk?
The focus depends on the audience. Leadership teams get a strategic perspective; technical teams can go deeper into retrieval, graphs and agentic workflows.
Do we need Neo4j or LangGraph in-house?
No. The talks explain architectural principles. Neo4j and LangGraph serve as concrete examples; you do not need to use these tools already.
What do you need for an inquiry?
Helpful details are your preferred date, venue or setting, audience and size, topic, and the language you want. Availability is checked case by case.
Your next step
Agree on date and focus.
We tailor examples and depth to your search problem and your audience. No prior knowledge of search algorithms is required.
Tell me about your audience or team, your question and a preferred date. On that basis, we clarify format and scope.
See the workshop on this topic