Dennis Irorere

Hey there. If you're meeting my curated digital self, well, nice to meet you. I hope to meet you properly someday.

A simple rundown. I'm interested in, and actively working on, representations, information quality, and the data engineering underneath it. Entity resolution, NER, classification, and digital identity: decentralised, self-sovereign, and the question of who decides your data is yours. In more marketable terms: record linkage, knowledge graphs, semantic layers, and intelligence.

Lately that means AI agents, and whether one understands what it is acting on, and knows when to refuse. That's me.

Data for the Rest of Us is out. Order a copy.

Latest

Note · rev. 24 Sep 2026

The generator is the dataset

Four waves of where data came from. The fourth one does not remove the hard problem, it moves it somewhere nobody is looking.

Note · rev. 13 Sep 2026

The name a system remembers

Canonicalization is not just tidying text. It is how an organisation records what a reference is meant to represent.

29 Aug 2026

What stays human

The human role is not to beat the machine at recall, calculation, or producing a first draft.

17 Aug 2026

Real Takes Time

A new site, two months of silence, and why I decided to lean in rather than out

Writing the journey, as it happened

All writing →

Notes living, revised

  • The generator is the dataset Four waves of where data came from. The fourth one does not remove the hard problem, it moves it somewhere nobody is looking. rev. 24 Sep 2026
  • The name a system remembers Canonicalization is not just tidying text. It is how an organisation records what a reference is meant to represent. rev. 13 Sep 2026
  • A category is a decision The same marketplace, address book, or neighbourhood can be grouped many ways. Somebody chose, for a purpose, and the purpose is still in there. rev. 9 Sep 2026
  • Similarity is not transitive Looking alike can help us find a group. It cannot tell us what the group is for. rev. 7 Sep 2026
  • Sameness and similarity Two records can look alike without referring to one thing. The small distinction that makes matching systems safer. rev. 3 Sep 2026
All notes →