Structured AI proofreads construction drawings the way spell-check proofreads a sentence - before the mistake becomes concrete.
Somewhere in a New York office, an AI is reading a construction drawing that a human would need a full afternoon and a strong coffee to review. It is checking every callout against the building code, cross-referencing the electrical set against the structural one, and noting - politely, in a sourced report - that a duct and a beam are trying to occupy the same cubic foot of a building that does not yet exist. This is Structured AI at work. It does not sleep, it does not skip page 214, and it never has a Friday-afternoon lapse in concentration.
The construction industry runs on documents almost nobody has time to read in full. Building codes stretch longer than most novels. A single project can generate thousands of drawings across a dozen disciplines, each one a chance for a small error to travel quietly from ink to steel - where fixing it costs a fortune. Structured AI's bet is unglamorous and, on inspection, obviously correct: the drawing review that engineers dread is not a judgment problem. It is a reading problem. And reading, at scale, is exactly what machines have gotten frighteningly good at.
Many of the great AI companies of the next decade are being built right now, and I want to be a part of that.
Raymond Zhao went to Oxford to study maths and statistics, convinced the natural next step was finance. Then he did the internship.
The Goldman Sachs summer was, by his own account, uninspiring - the sort of experience that quietly reroutes a career. Zhao walked out of finance and toward artificial intelligence, joining Oxford's venture capital society along the way. That society did two useful things: it introduced him to a network of founders and investors, and it introduced him to one of his eventual co-founders.
What emerged was Structured AI, co-founded in 2025 with Issy Greenslade and Brandon Abreu Smith. The company picked a market most AI founders avoid precisely because it looks boring from the outside: architecture, engineering, and construction. Boring, it turns out, is where the money and the missed deadlines live.
The pre-seed came together fast - a reported £375k in about a week - and Y Combinator followed. The thesis stayed constant from pitch to product: give construction an AI workforce, and start it on the job everyone agrees is necessary and nobody enjoys.
Oxford maths & stats; ex-Goldman Sachs intern turned AI founder.
Part of the founding trio building the construction AI workforce.
Co-founder helping turn drawing review into an automated system.
The platform is a set of agents, each aimed at a task engineers currently do by hand, on deadline, hoping they did not miss anything.
Reviews drawing sets against building codes and standards, returning a deterministic, sourced compliance report you can actually defend.
Ask a drawing set questions in plain English. Answers come back cited to the exact source page.
A no-code builder for AI agents that enforce your firm's own templates and standards - your rules, automated.
Finds cross-discipline conflicts with auto-aligned overlays, catching the duct-meets-beam moment in ink, not on site.
Queries live BIM models for element-level QA/QC, inside the design tools engineers already live in.
Keep the human in charge of the call. Take the tedium off their desk. That is the whole design philosophy.
Top 1% of professionals I've worked with. Their AI solutions and proactive communication have transformed our Innovation Program.
Structured AI is already inside architecture, engineering, and construction shops - the people who feel the pain of a missed drawing note most acutely.
From a pre-seed that closed in roughly a week to a $4.2M seed a year later - the kind of curve that makes investors lean forward.
Pre-seed: Zero Prime Ventures, Airtree Ventures, Oxford Seed Fund. | Seed led by FCVC with Y Combinator, 20VC, Cherry Ventures, Zero Prime Ventures, Transpose Platform & Sequoia Scout. Bars scaled to round size.
Product demos and founder interviews aren't consistently published yet - the links below go to the primary sources and channels. Search the company's YouTube and X for the latest walkthroughs.