What I have built and what it proved.
Vistafy: rendering and motion inside the practice.
Architects turn sketches, references, and finished renders into client-ready visuals and walkthroughs without leaving the project workspace. The lesson is that AI is more useful when it holds context across iteration, review, and presentation.
BookDone: voice and photos into apprentice bookwork.
Site work gets captured in the way tradespeople already work: photos, voice notes, and short follow-up questions. The system turns that field evidence into formatted BCITO submissions that still need review before they count.
Buildbit: NZS 3604 answers with the clause attached.
NZS 3604 became askable in plain language, under a formal Standards NZ license, with answers tied back to clauses. The useful lesson was also the boundary: in compliance work, almost-right is still a liability.
NZ site intelligence: land context before design hardens.
Parcel, terrain, constraints, and missing-data states get pulled into an early feasibility layer. The aim is a practical source pack that helps a team decide what needs review before concept work goes too far.
> set objective > agent planning... > agent 47 / 100 > APPROVE → continue > next task
Command Center: agent work behind approval gates.
A web dashboard for running AI agents like a team. Set an objective, the agents plan and execute, and every plan and result waits on human approval before it counts.
Parallel: corrections become memory.
A native desktop app that drives your local coding agents, Claude Code, Codex, and Cursor, from one place. Actions wait on approval before they run, and every correction is carried into the next run.
Where research turns into tools you can open.
Designing with AI.
Sculpt a building's massing by hand, then two models turn that drawing into a concept render and a 3D model you can move around. The experiment is the interface: designing with AI rather than prompting it.
Prompt to 3D.
Describe a house in a sentence, generate a concept image, then lift it into a 3D model you can move around. Two models run in sequence, end to end in the browser.
Market Pipeline Watch.
It turns public demand signals, an example firm baseline, and source gaps into a short market snapshot a practice owner could read, with the gaps named rather than papered over.
Each piece of research hangs on one clear question.
What lets AI do the real work while the people in charge stay responsible for it?
A source-backed read on where AI is genuinely changing architecture practice: the market shift from features to workflow control, the operating layer AI has to be onboarded into, and the control loop three prototypes kept arriving at.
Read the report → The first hour on a site · Research noteHow can AI strengthen the judgement a person brings to a site?
Assembling a site from public records into one readable layer, the two ways to build that layer, and where it sharpens a professional's judgement instead of replacing it.
Read the note → A price you would sign · Research noteCan AI produce an early cost view a quantity surveyor would stand behind?
Early estimates decide whether a project goes ahead and are made on thin information. A source-backed look at how AI can speed the number up while the quantity surveyor still owns it, grounded in New Zealand cost data.
Read the note →