Skip to content

What we learned
building these.

Not marketing. These are the findings that changed what we built — the evidence on how people actually learn, what it takes to trust a model’s output, and how much of a building site you can measure before anyone visits it.

Learning science8 min

The learning techniques that work don't need a screen

A synthesis of 37 sources on what actually raises attainment — and why almost every technique with a real effect size works at least as well on paper and by voice as on a display.

Read
AI engineering7 min

Generating questions is easy. Verifying them is the product.

Any current model will write you a plausible multiple-choice question. Getting one that is provably correct, with exactly one right answer, is a different engineering problem — and it is the only part that matters.

Read
Architecture technology9 min

A site survey assembled entirely from public data

Terrain, sun path, a decade of climate, three decades of flood history, seismic exposure and what the neighbours overshadow — all obtainable before anyone visits the plot, and most of it free.

Read
Architecture8 min

What “enterprise-ready” actually means

Turning a consumer product into one an institution can buy is not a feature list. It is a boundary — and almost everything hard about it follows from deciding where that boundary sits.

Read
AI engineering7 min

Design-to-code is easy to demo and hard to ship

Converting a design into markup takes an afternoon. Converting it into a codebase that compiles, is responsive, and a developer will accept is a different problem — and the gap between them is where most tools live.

Read

Want this thinking on your problem?

Book a free 30-min call