The physical world generates enormous amounts of data that almost nobody can actually read
Here is a fact about commercial real estate that should bother you more than it does: a great deal of it is still surveyed by hand. Someone drives to a building, looks at the roof, writes a report, and by the time the report reaches the person making a nine-figure decision, the roof may have changed, the zoning may have changed, and the report is - in the technical sense - old. Geolava's co-founder and CEO, Hantz Fevry, puts it plainly: "Major investment decisions are being made basically on reports that are most of the time obsolete." That is the whole business in one sentence.
Geolava, a San Francisco startup that came out of the usual mix of stealth and seed funding, wants to fix the obsolescence problem by building what it calls a world model for the built world. The idea, roughly, is to take every kind of picture you can take of a place - optical, thermal, LiDAR, spectral, street-level, synthetic aperture radar - and fuse it with the boring but load-bearing paperwork of real estate: zoning rules, land-use constraints, tax exposure, deed history. Then you point a foundation model at the whole pile and let people ask it questions in plain language.
The elevator version is "Google for the physical world," which is the kind of phrase that sounds like a cliche until you notice how much of the physical world is genuinely un-searchable. You cannot type "which buildings in this portfolio have a roof that is about to fail" into anything and get an answer. Geolava would like you to be able to, and it would like to be the thing you type it into.
What makes this more than a demo is the team's insistence that the unlock is not more raw data. Everyone already has too much data. The unlock is a model that understands what the data means - that can look at a thermal signature and infer an HVAC upgrade, look at a roofline and rank it against the rest of your holdings, look at a parcel and flag a zoning problem before the city does.
We bring AI to the physical world.
Four verbs, one intelligence layer
Geolava organizes its product around four things institutions do to a property - and tries to make each one take a sentence instead of a site visit.
Analyze
Assemble a full property profile from fused imagery and public records. Compare roof quality across a portfolio, spot a water view, measure proximity to risk.
Underwrite
Develop and validate an investment or insurance thesis with real-time signals on value, condition, tax exposure and regulatory risk - not a stale PDF.
Monitor
Track how an asset changes over time. Detect construction, HVAC upgrades, decay and zoning violations as they appear, not at the next inspection cycle.
Forecast
Simulate how a property evolves across a 5-10 year horizon, so investors can act before the rest of the market has noticed anything is happening.
Seven kinds of vision, one language
Most AI startups chase the screen. Geolava chases the street. Its foundation model treats these very different ways of seeing a place as one language for describing physical reality:
- Optical satellite imagery - the picture you already imagine
- Thermal - heat signatures that hint at systems and use
- LiDAR - three-dimensional structure and elevation
- Spectral - material composition beyond visible light
- Synthetic aperture radar - sees through cloud and dark
- Street-level imagery - the human-eye view
- Government records - zoning, land use, tax, deeds
Where the value sits
Illustrative emphasis based on public positioning - not a financial disclosure.
The intelligence layer for institutions to perceive, understand, and predict physical reality across space and time.
The founders who mapped 130 million buildings - and then started over
Hantz Fevry and Pierre Frederic Mombeleur are both Haitian-born and met at Google. Their previous company, Stoovo, was a location-intelligence platform that mapped more than 130 million buildings and was acquired by Doorstep in 2024. Fevry's resume runs through Google, DeepMind, and an MIT specialization in quantum computing - which is a lot of horsepower to point at the question of whether a roof is any good.
Hantz Fevry
Repeat founder (Stoovo). Ex-Google / DeepMind. Moved to the U.S. in 2009, studied at Stony Brook, completed an MIT quantum specialization.
Pierre F. Mombeleur
Co-built Stoovo's location-intelligence platform. Met Fevry at Google; fellow Haitian native and long-time collaborator.
Andrew Cutler
Listed among Geolava's founding team building the spatial foundation model and platform.
A short, deliberate cap table
Investor mix reflects the built world it serves: a fintech fund, a construction-tech arm, and a space-tech backer.
Notes from the margin
- Space to street. The team frames its work as translating "spatial physics into financial advantage."
- Pedigree. Stoovo, their prior company, counted Apple, GM and FedEx among users.
- Quantum detour. Fevry completed an MIT specialization in quantum computing before all this.
- Lean. Roughly 14 people are aiming at a $20T+ market.
- The bet. Not more data - a model that understands the data you already have.
Owners, lenders, insurers - and anyone tired of waiting for the report
Geolava sells to institutions that touch physical assets: real estate owners, investors, asset managers, brokers, lenders and insurers underwriting or monitoring property. Applications reach into insurance underwriting, climate and regulatory risk, site selection and portfolio monitoring. Competitors and alternatives include geospatial and property-intelligence players like Cape Analytics, Zesty.ai, Nearmap, Regrid and EagleView - and, more stubbornly, the manual surveying that Geolava is trying to make optional.