Depixen

Depixen was founded on a belief: that knowledge an AI system can actually rely on is one of the rarest things in the world right now.

The models will get better every year. What's surprising is that the data those models are meant to reason over is still scattered, unverified, and disconnected from itself. That isn't a model problem. It's a problem with the knowledge itself.

Our work is to rebuild that knowledge. A layer that ties together an industry's products, decisions, standards and actors, where every record carries where it came from, who verified it, and when. An ontology isn't a technical detail. It's the bridge between where people decide something and where machines can act on it. Over the next decade, whether an AI system can be trusted will depend less on its model than on how that bridge was built.

When we set out to build this layer, we realised the data itself didn't exist. No verified product source. No record of the decisions happening between people in the sector. No machine-readable representation of how work gets done. So we had to build the data space too. ArchiPick is that space. It brings together three things on one graph. The interactions between people. Verified product data modelled against normative standards. The workflows we lifted out of manual processes. Together they behave like an operating system for the construction industry.

We didn't build this in a lab. Over 200 organisations worked on the pilots with us. We sat down with manufacturers and modelled their products with them. We ran joint research with universities. We put our own money into R&D.

Depixen is a member of the W3C, the body that defines RDF, OWL, SPARQL and PROV-O. We're now preparing joint pilots with AI companies, to see how this layer behaves when a general-purpose model runs against it.

Depixen is based in London. Series A closed at $5M.