On a construction sheet marked M-401, a mechanical engineer traces a single door tag. It points to a schedule three sheets away, which points to a specification section, which references a submittal, which is contradicted by a bulletin issued last Tuesday. Multiply that by a thousand sheets. This is the daily reality of building anything large - and for decades, no software actually understood it. It just stored the PDFs.
Primepoint is the company betting that this - the reading, not the storing - is the real problem. Its AI ingests plans, specs and schedules and treats them as one interconnected system. It links door tags to schedules, submittals to specs, RFIs to drawings, automatically, down to the tag level. The output isn't a search result. It's a knowledge graph of the whole project.
The person leading it is not a construction lifer. Lubomir Bourdev was a founding member of Facebook AI Research. He built the object-recognition system that ran on every photo and video posted to Facebook and Instagram. His next act, a deep-learning video-compression startup called WaveOne, was acquired by Apple in 2023. He holds more than 100 patents and carries north of 100,000 academic citations.
So when Bourdev decided the hardest remaining computer-vision problem was hiding inside a mechanical plan, people with money listened. In April 2026, Primepoint closed a $10M seed round. Among the backers: Yann LeCun, a Turing Award winner and one of the founders of modern deep learning, writing an angel check.
