Knowledge graphs,
built to be used.

Practical Neo4j and graph-AI work - ontology design, multi-source ingestion, and retrieval that holds up in production. By Konrad Kaliciński.

// context graphs - grounding LLM answers(:Document)-[:MENTIONS]->(:Entity)-[:GROUNDED_IN]->(:Context)
// supply chain - component traceability(:Supplier)-[:SHIPS]->(:Component)-[:ASSEMBLED_INTO]->(:Product)
// crop protection - biorationals against pests(:Biorational)-[:TARGETS]->(:Pest)-[:THREATENS]->(:Crop)
// legal - case law over regulations(:Ruling)-[:INTERPRETS]->(:Article)-[:PART_OF]->(:Regulation)

01 - ABOUT

I build and ship knowledge graphs on Neo4j - ontology design, multi-source ingestion, GraphRAG, and the unglamorous plumbing that makes a graph survive contact with real data.

I write about this work every week: what is working, what is not, and what most teams get wrong on their first knowledge-graph project.

- Knowledge graph practitioner and writer: building, teaching, and consulting on graph-AI projects.

02 - WRITING

Medium LONG-FORM · WEEKLY Long-form articles on Neo4j, GraphRAG, and applied graph-AI.

RECENT ARTICLES

How I Test a Knowledge GraphCompetency questions and invariants turn "the graph looks right" into something you can verify - shown on a 10-K filings graph anyone can read.Do Knowledge Graphs Fail as Agent Memory? What Two 2026 Benchmarks Actually ShowTwo studies say graph memory loses on cost and accuracy. A closer read - plus the memory tools worth trying, for coders and non-coders alike.How Do You Know That? Adding a Provenance Layer to a Supply-Chain Knowledge Graph in Neo4jA traceability graph says where a batch came from. It usually can't say how you know - and that gap is a modeling choice, not a tooling problem.The Supplier That Stops Everything: Finding Supply Chain Chokepoints with Neo4j and GDSToyota paused hybrid SUV orders over a part four tiers deep. BASF shut ammonia plants that fed four industries. Both are the same graph problem.Beef Lasagne, Five Countries, One Graph: The Horsemeat Scandal and Neo4jA frozen lasagne crossed five countries before anyone could name what was in it. The fix wasn't more paperwork - it was modelling the supply chain.From Spreadsheet to Substitution Engine: Building an Agricultural Knowledge Graph in Neo4j, Layer by LayerBuild an agricultural knowledge graph one public data source at a time, and watch the questions it can answer get more interesting with every layer.Agriculture Already Has Its Ontologies. The Knowledge Graph Is Where They Expire.The ontology is the easy part. Keeping its facts from going stale is the whole job.
LinkedIn SHORT TAKES Shorter takes and project notes from ongoing graph work.

03 - SELECTED WORK

Polish plant-protection product register →

A graph of registered pesticide products, active substances, crops, target pests, application windows and dosage - built from official register data and product labels, with a working question-answering demo.

SEC 10-K Filings Graph →

25 annual reports from 5 mega-cap companies as one graph of relationships - competitors, dependencies, and risks tracked over time - built from SEC 10-K filings with a working question-answering demo.

More verticals in progress - new graphs land here as they ship.

04 - CONTACT

Working on a graph?

I take on consulting engagements, knowledge-graph reviews, and collaboration on graph projects. Send me a few lines about your graph on LinkedIn - I reply to every serious message.

Message me on LinkedIn