Data Architecture & Modeling

Modern data architecture has accumulated layers - catalogs, glossaries, dictionaries, metric stores, lineage tools, observability platforms - each modelling a slice of meaning in its own format. The result is fragmentation that breaks the moment you connect an AI agent: every layer disagrees on what "active customer" means.

This collection rethinks data modelling for the AI era. Posts examine how the typed semantic graph collapses scattered metadata into a single, versioned source of truth; why the 20-year deadlock between dimensional models, data vaults, and lakehouse formats is broken not by another modelling philosophy but by elevating semantics above the physical layer; and how accidental complexity in the modern data stack compounds when every tool encodes meaning independently.

You'll also find architectural deep dives on the enterprise memory graph, the autonomous patterns that maintain it, and the cost equations behind building your own semantic execution layer versus consuming one as governed infrastructure. The thesis across these pieces: the next generation of data architecture is semantic-first - meaning is modelled once, in a typed and reasoning-friendly form, and every downstream consumer compiles through it rather than rebuilding it.