sparkling·ideas

Henrik Pettersen · London · Ontology & Agentic Engineering

Shared meaning,engineered.

Organisations don't struggle to store data — they struggle to agree what it means. For 25 years I've built the ontologies, knowledge graphs and standards that carry meaning across organisational boundaries. Now I deliver them with agentic engineering: AI that drafts, extracts and maps — under an engineer's judgement.

Start a conversation See the practices ↓

valid Turtle · the rest of this page is the documentation

Two practices, one loop

Practice · Semantics

Ontology & knowledge graphs

The models that make data mean the same thing everywhere — ontologies, taxonomies, vocabularies, and the validation that keeps them honest.

  • Built an OWL database in 2002; contributed to the early SPARQL specification
  • Led engineering on SNOMED CT — clinical terminology in production across 45+ countries
  • Modelled data for 7 ISO standards at the European Medicines Agency
  • Linked-data platforms for UK government intelligence, banking, retail and energy
  • Currently: the linked-data model for the UK property-data standard (PDTF)

OWL · RDFS · SKOS · SHACL · RML · SPARQL · Datalog · Neo4j · RDFox · QLever · Neptune · MarkLogic

Practice · Agents

Agentic engineering

AI agents changed what one engineer can deliver. I run agent-first delivery: pipelines that ingest sources, extract schemas, draft mappings and generate queries — with tests and human review deciding what ships.

  • Claude Code as primary delivery tool across four consecutive contracts
  • Trained an entire engineering team in agentic workflows at Telicent
  • Agent frameworks for RML mapping, schema extraction, SPARQL generation and GraphRAG
  • Multi-agent council method: evidence in, challenged drafts out, humans ratify

Claude Code · MCP · multi-agent orchestration · GraphRAG · AWS · GCP · Terraform · Kubernetes · Kafka

Why both

AI→ontology→better AI

The practices compound. Agents accelerate the modelling — turning documents, schemas and expert interviews into challenged, testable drafts. The governed model then grounds the next generation of agents: explicit vocabulary, relationships, rules and provenance instead of prompt guesswork. That loop is the offer.

The record

25 yrssemantics & data standards
45+countries running SNOMED CT
7ISO standards modelled at the EMA
2002my first OWL database
SNOMED International✦European Medicines Agency✦HSBC✦Telicent✦H&M✦Kaluza✦OVO Energy✦Pearson✦DeFaqto✦Eurocamp✦TUI✦Freshfields✦Amnesty International✦Boon Global✦

Contact

Have a data-meaning problem, or an AI-delivery ambition?

Available for consulting and contract engagements — ontology strategy, knowledge-graph delivery, agentic-engineering adoption, or all three at once.

henrik@sparklingideas.co.uk LinkedIn GitHub