I have lived in the Bay Area for nearly twenty years, most of that time in San Francisco. My wife and I were married here and our son was born here. It is the city I know best, and it is a city that has rebuilt itself more than once.
I have spent about as long working in technology — twenty years, most of it as a product designer, and more recently building with machine learning and inside AI startups. That combination is the reason BuiltAtlas takes the shape it does. I design the thing and I build the thing, which means the questions come from what a person actually needs to know, and the answers have to survive an engineer reading the source.
Architecture and history have held my attention for as long as I can remember. Not as a profession — as a persistent curiosity. I would walk past a building and want to know who designed it, what stood there before it, why it was built the way it was, and what had happened to it in the century since.
The answers almost always existed. They were simply scattered: a designation in one registry, a permit in a municipal file, an architect's name in a monograph nobody has opened in forty years, a photograph in an archive. Any single question was answerable. Asking a hundred of them was not.
The record exists. It has simply never been assembled.
I started BuiltAtlas assuming I was building it for people like me — curious, on a sidewalk, wanting to know what they were looking at.
Then I looked at what professionals do with the same information, and found them standing at the same wall with a great deal more at stake. An architect handed an existing building spends days assembling that scattered record before anyone is paying them. A developer deciding whether an office building can become housing screens on partial information, because complete information is expensive. Only twenty to forty percent of buildings turn out to be viable conversion candidates — and the rest are discovered slowly, and at cost.
That range is a published industry estimate, not BuiltAtlas data. We will say so every time we quote a number we did not produce.
The same wall. Very different stakes.
What we are building
BuiltAtlas is the knowledge layer for the real world. That is the long ambition. The thing we sell today is narrower and more useful: an intelligence layer for existing buildings, so professionals can weigh constraints, history, risk and opportunity before making expensive decisions.
It resolves a building — from a photograph, an address, or a parcel number — into one structured record: who designed it, what it is made of, how it has been altered, what regulates it, and what can reasonably be done with it next. Every field carries its source, its date, and whether a named authority asserted it or BuiltAtlas derived it from inputs we name.
Buildings are the first entity type. They are the most documented, the most valuable, and the most consequential thing in the built world. They will not be the last.
We are working in San Francisco, New York and Seattle.
How we work
Every field says whether a named source asserted it or we derived it, and from what. No invented confidence scores — the number is auditable instead.
Every record lists what has not been sourced. A blank field means not yet ingested, never zero. A record that hides what it does not know is worse than no record.
Public records contradict each other constantly. Where sources conflict we show the conflict and both sources, rather than resolving it quietly in the background.
Who we serve
Working with us today
Teams working on existing buildings — adaptive reuse, renovation, and additions — who need existing conditions, permit history and regulatory context before a project is even won.
Investors and developers screening buildings for conversion or repositioning, who need to separate the viable from the dead ends before spending on full diligence.
Consultants and heritage specialists assembling designation status, district context and alteration history for older buildings.
Where the same record goes next
Underwriting existing assets on construction type, age, retrofit obligations and alteration history rather than on what a borrower reports.
Connecting parcels, zoning, permits, designation and development into a single queryable picture of a district.
An MCP server and API so an agent can reason about a physical place from a sourced record instead of scraping the open web.
Anyone who has stood in front of a building and wanted to know what they were looking at. That is where this started, and it is still in the product.