ViableSite is more than a feasibility platform. We are building the model (and we are the rare team positioned to build it) that reads any site the way a professional does, and turns fragmented land data into one durable record the industry can trust.
Every site has a physical truth: what sits above the ground, what runs below it, and what can realistically be built. Today that truth is read from scratch on every project, by a different team, and thrown away each time.
The hardest part is not reading any single source. It is producing a conclusion a professional can stand behind, with honest confidence, when the ground truth that would verify it is scarce and expensive to collect. Underground risk is the sharpest case: the national utilities record is restricted, surveys are costly, and a wrong call is expensive.
We are building physical-site inference for the built environment: a multimodal model that reads a site across several data sources at once and produces an understanding a professional can act on, with honest confidence attached.
A green patch on a satellite image means little on its own, but cross-referenced with a historic map showing a road once ran through it, it becomes a flag: something was there, and something may still be below. Reading a site the way a professional does means holding those sources together and reasoning across them.
The hard part isn't reading any one of them. It's knowing how much to trust the conclusion when the ground truth that would confirm it is scarce. That problem, calibrated prediction under sparse, expensive ground truth, is the heart of what we're building, and it's what separates a model the industry can rely on from a confident guess.
Every output is tested against real-world ground truth and signed off by a chartered architect before it ships. Construction does not act on a black box - calibrated confidence plus chartered validation is what makes a model output something the industry will actually use.
A durable record of the built environment can only be built from inside the workflow that creates the labels: a real site, a real assessment, validated by a chartered professional. A large AI lab has the technology but not the workflow. A proptech tool has neither the chartered discipline nor the underground-risk problem to solve.
The data sources themselves aren't secret as most can be licensed or accessed by anyone willing to assemble them. What's hard is the discipline of turning them into a conclusion a chartered professional will put their name to, every time. That's the work, and it's work we're already doing.
Standing in that junction takes three capabilities that rarely sit in one team: deep built-environment expertise, production machine learning on physical-world data, and academic research standing. We have all three.
18 years across £43M–£250M projects (PRP, Aedas, KAR Studio). Led Oxford City Council's Small Sites Programme — 256 sites assessed, 435 homes identified, 70 enabled. The deep built-environment expertise that grounds the model in how professionals actually read a site.
20 years in software and cloud infrastructure. Built satellite-verification systems at Sylvera (carbon credits) — physical-world inference at scale. Python · TensorFlow · cloud.
8+ years of AI in construction; contributor to UN and EC AI policy. The research standing to build and validate the model credibly.
A cross-disciplinary group of chartered built-environment professionals — covering every core discipline that reads a site: planning, surveying, architecture, civil engineering and structural engineering. They pressure-test the methodology against how each profession actually works.
Selected 1 of 8 for the 2026 cohort (Harwell · SatApps Catapult).
Scientific Lead is a Fellow.
Methodology selected.
Women in AI winner.
Recognition through ESA BIC UK at Harwell.
We're hiring, partnering, and raising. If you work on calibrated AI, geospatial inference, or the built environment — we'd like to talk.