Every project leaves a paper trail. We’re trying to turn it into a memory.
PDFs, scans, and vector drawings were designed for human interpretation, not machine understanding. We read them visually, infer intent from conventions, and mentally connect symbols to schedules, notes, and details. Software, on the other hand, sees lines, text fragments, and coordinates — but no meaning.
“The construction industry doesn’t lack data — it lacks models that understand what the data actually represents.”
This is where foundation models come in. In other domains, foundation models have transformed raw inputs — text, images, audio — into structured, reusable intelligence. But construction documents pose a unique challenge: they are multimodal, highly contextual, and governed by decades of domain-specific conventions.
Why General AI Models Fall Short
A general-purpose vision or language model can extract text, detect shapes, or describe what it sees. But construction documents require something deeper. A wall is not just a rectangle. A tag is not just text. A dashed line does not always mean the same thing.
Understanding drawings means understanding relationships: how a door relates to a wall, how a space relates to a program, how a detail relates to a callout, how revisions propagate across sheets. These relationships are implicit, not labeled — and they vary across firms, regions, and disciplines.
“You don’t train a construction model by showing it images — you train it by teaching it how projects actually work.”
What a Foundation Model for Construction Really Means
A foundation model for construction documents is not a feature. It is an intelligence layer — one that learns the grammar of drawings, the semantics of symbols, and the logic of coordination.
At its core, such a model must:
- Understand drawing conventions across disciplines and regions
- Identify objects, not just geometry
- Infer relationships between elements, sheets, and references
- Track changes and revisions as first-class information
- Convert visual intent into structured, queryable data
This intelligence does not replace BIM, CAD, or existing tools. It complements them — acting as a bridge between legacy documents and modern digital workflows.
From Documents to Systems
When construction documents become machine-understandable, they stop being endpoints. They become systems — systems that can be queried, validated, compared, and reused.
Quantities become traceable. Conflicts become detectable before they reach the field. Knowledge from one project informs the next. The drawing set evolves from static record to operational intelligence.
“The future of construction is not more documents — it’s fewer surprises.”
We’re building this because the people who build the world deserve tools that truly understand it.