01
Describe search from the user’s perspective
Collect typical questions, hard cases and useful answers; phrase expectations so they can be tested.
Workshop / Search & Retrieval
BM25, vector or graph? The right search for your use case.
We translate your search problem into concrete questions and comparison criteria. The result is a well-founded architectural approach: Where is lexical search enough? Where does semantic context help? Which relationships would need to be explicit for a graph?
1 or 2 daysGerman / English1 day from €6,500 / 2 days from €12,500 plus VAT
Who it’s for
For teams with a concrete challenge.
For search, data and AI teams as well as business owners with a concrete search or retrieval task.
What the workshop produces.
- A first question set with expected results and transparent relevance criteria.
- A decision matrix for BM25, vector search, graphs and possible combinations.
- An architecture sketch and a scoped validation plan with open questions on data and access.
Content framework
From understanding to a working artifact.
We tailor the focus and technical depth to your question.
02
Compare approaches against the questions
Examine which terms, meanings or relationships an answer requires and which data is available for it.
03
Make the decision testable
Narrow down candidates, define quality checks and sensibly scope the next technical experiment.
Your starting point becomes working material.
If possible, bring typical search queries and approved sample content. Whether a hands-on search comparison is part of the workshop depends on your data and existing infrastructure.
The listed deliverables define the intended scope of work. Scope and technical exercises are set out in the proposal before the workshop.
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.
What does a workshop cost?
1 day from €6,500 / 2 days from €12,500 plus VAT. All fees plus VAT. The exact scope of work is set out in the proposal.
What do we take away from the workshop?
Depending on the topic, you get a use case canvas with a roadmap, a first graph model with a validation brief, or a state diagram with evaluation cases and a prototype scope. Each topic page lists its deliverables.
Is the workshop remote or on-site?
Both formats are possible. Working mode, technical requirements and scope are tailored to your team.
How large should our team be?
For the Machine-Readability Check, the setup is the head of IT plus two to three business unit leads. All other topic workshops are designed for up to eight participants.
Do we need a large data platform already?
No. We can start with a narrowly scoped domain, existing documents or a specific process. The starting point depends on where you are today.
Do we need Neo4j or LangGraph licenses?
These tools are not required for conceptual work. Whether a hands-on technical exercise makes sense, and what access it needs, is clarified in advance.
How do we handle confidential information?
Please do not send confidential documents with your first inquiry. Access, suitable examples and any required non-disclosure agreement are settled before we start working together.
Are the intro call and workshop available in English?
Yes. I speak fluent English. All offerings and intro calls are available in German and English.
Your next step
Agree on date and focus.
If possible, bring typical search queries and approved sample content. Whether a hands-on search comparison is part of the workshop depends on your data and existing infrastructure.
Tell me about your audience or team, your question and a preferred date. On that basis, we clarify format and scope.
See the keynote on this topic