2D floor plans read and rebuilt as 3D BIM models
A computer vision engine that reads floor plans and builds editable 3D models, flagging anything it isn't sure of.
- Software and AI product
- Live, accuracy being measured
At a glance
Client Buildvick Engineering
Client
Buildvick Engineering
Industry Construction technology
Industry
Construction technology
Time to build 12 weeks
Time to build
12 weeks
Team 4 specialists
Team
4 specialists
Year 2025
Year
2025
Status Live, accuracy being measured
Status
Live, accuracy being measured
Every case on this site answers the same four questions, so you can compare them.
Before
Architects spent days tracing 2D plans in CAD to build 3D concepts for pitches, which slowed every sale.
What it does now
A plan is uploaded and cleaned up for reading.
The engine finds walls, doors, windows and wall intersections.
Wall segments become coordinates and a connectivity map.
A 3D model is built in the browser, where heights and materials can be changed.
The model exports to standard formats, including OBJ and DXF.
Where a person decides
Every element the engine isn't confident about is flagged. A person checks it before the model is used.
Result
We're measuring precision and recall on drawings the engine has never seen, with a human BIM modeller as the judge. We'll publish the figures when that run is complete.
How we built it
Built image pipelines that clean and sharpen plan uploads.
Trained a model to detect doors, windows and wall intersections.
Wrote the geometry that turns segments into a connected floor plan.
Built the 3D view with Three.js and WebGL, and an editor for heights and materials.
Added export to standard BIM formats and an evaluation harness for accuracy.
Stack
- Python
- OpenCV
- FastAPI
- React
- Three.js
- WebGL
- TypeScript
From the project
"BlueprintAI changed our client pitches. We can now upload a 2D PDF drawing on a call and show the client a full 3D interactive model of their floor plan in minutes. It has cut days out of our design cycle."
More work like this
A 3D configurator that turns visitors into quote requests
RealtyOS: a live booking board that ended double-bookings
Big tasks broken into small, timed steps for ADHD
Your case study starts with an audit.
One to two weeks inside your operations. You get a map of where time and money leak, the return on each fix, and a fixed scope before we build anything.
Most businesses we meet have paid for an AI tool nobody uses. We start from your numbers, so the first build is chosen by payback.
Zain Khan and Anwar Sheikh
Founders, OutGrow
Outgrow
Intelligence
Studios.