Skip to content

Floor plan AI

Floor plan AI, fine-tuned to your drawings

Our models are already trained to read walls, rooms, doors, windows and symbols off a floor plan. We fine-tune them on the drawings you actually receive, for the fields you actually use, and hand back structured data your software can query.
Analyse my plan, freeSee how it works
  • PDF, DXF and scanned plans
  • Rooms, counts and areas as JSON
  • Runs in your cloud or on-prem
Residential floor plan with each room detected, colour-coded and labelled with its dimensions and area, beside a room list totalling 88.96 square metres
Our models on a residential plan, before any tuning: rooms found, measured and listed.

Why it is hard

A drawing shows a room's area. It never states it.

The area is implied by two wall lines and a scale bar. The room's name is loose text floating somewhere near the middle. A door's width is in a schedule three sheets away. Every number you need is on the sheet and none of it is in a field anything can query, so somebody reads the picture and types it in.

  • Text at odd angles

    Tiny, rotated to follow walls, overlapping hatches. Ordinary OCR misses it.

  • Symbols that look alike

    Abstract line art that changes between firms, regions and decades.

  • Walls drawn many ways

    Double lines, fills and hatches all mean “wall”. Openings are just gaps.

  • General AI guesses

    Ask a general model to count doors and it will answer confidently, and wrongly.
TopicTracing by hand, or a general AI modelA model tuned on your drawings
SpeedHours of tracing and measuring per sheetMinutes per sheet once the pipeline runs
MeasurementsEstimated by eye, or invented by the modelComputed from geometry at drawing scale
OutputA marked-up PDF, or a paragraph of proseStructured JSON with rooms, openings and adjacency
CheckingRe-measure the sheet to find the mistakeEvery value links back to where it was found

Where every project starts

What our models already read

These are our base models: models we have already trained on floor plans, as opposed to general document AI pointed at a drawing. Six layers, shown on real commercial and residential sheets, with no tuning for any particular client.
  • Commercial floor plan with detected wall runs highlighted in different colours by wall type label

    Walls

    Wall runs are traced and grouped by the type label on the drawing, so partitions and exterior walls can be measured separately.

  • Room label read from a plan and the matching rooms shaded as polygons on a commercial floor plan

    Rooms and areas

    Room tags like “Restroom 106” are read, and the enclosed space around each tag is filled in as a polygon with an area at drawing scale.

  • Restroom area of a commercial plan with each detected door boxed and colour-coded by door tag

    Doors

    Door symbols are detected with their swing and grouped by tag, ready to match against the door schedule for widths and fire ratings.

  • Plan with window segments detected along an angled exterior wall and highlighted by window label

    Windows

    Windows are found along straight and angled walls and grouped by their labels, including long storefront runs.

  • Symbol template G1 beside a classroom plan where every matching tag is boxed and labelled

    Tags and symbols

    A template symbol, such as a G1 tag, is searched across the sheet, and every match is read, located and counted.

  • Plan where regions sharing the same hatch pattern are recognised and shaded green

    Hatch patterns

    Hatched and patterned regions are recognised, so areas can be separated by finish or material.

On drawings of a kind the models have seen before, this gets most of the way there on its own. It is rarely all the way there — which is the next section.

How we work

We fine-tune those models on your drawings

Nobody's drawings look like anybody else's — your symbol legend, your abbreviations, your title block, your scan quality. Fine-tuning a model that already understands floor plans takes weeks. Training one from nothing takes many months and a far larger labelled set, and that is the part you are not paying for.

  1. Step 1: Run our models on your sheets

    about 2 business days

    Send one plan and see what the models find on it before any tuning. It is the cheapest way to find out whether what you want is feasible at all.
    • No call needed
    • NDA on request
    Send a sheet

  2. Step 2: Fine-tune on your drawing set

    4–6 weeks

    We label a sample of your sheets and tune the models to your conventions, measured against your own manual results on fields you choose.
    • Accuracy target agreed up front
    • You own the trained weights
    How fine-tuning works

  3. Step 3: Run it where the work happens

    Ongoing

    An API inside your product, a batch job for your estimators, or a service in your own cloud. Retrained as new drawing styles arrive.
    • Low-confidence values flagged for review
    Integration and deployment

What you get back

Spatial data, not a picture of it

A structured description of the floor that your software can query. The schema is shaped around your fields on every project — this is the default.
  • JSON or CSV per sheet, by API or in batches
  • An annotated overlay, so a human can check it
  • Low-confidence values flagged, never filled in silently
More on floor plan analysis
Default fields in the floor plan analysis output
FieldWhat it contains
rooms[]Name from OCR, polygon, net area, adjacent rooms and the doors between them
walls[]Start and end points, length at drawing scale, type label
doors[]Tag, position, width, swing, the two spaces it connects
windows[]Tag, position, width, host wall
text[]Every string read on the sheet with its bounding box: dimensions, notes, room names
symbols[]Class, position and counts per room for fixtures and MEP symbols

Delivery

It has to land where the work already happens

A model that only runs on our machines is not much use to an estimator on a deadline. Drawings that cannot leave your network do not have to.

  • An API in your product

    Send a sheet, receive JSON. For PropTech and construction software teams.

  • Your cloud or your servers

    Deployed in your own account or on-prem when plans cannot leave the building.

  • Tuned to your sheets

    Trained on the symbols, fonts and drawing styles you actually receive.

  • A human in the loop

    Low-confidence items go to review, and corrections feed the next round of training.

Who we build for

One pipeline, different questions asked of it

The same structured output feeds a takeoff, a code check and a listing. What changes is which fields matter.

Start here

See it on your own drawing first

Reading about accuracy on someone else's plan tells you nothing. One of your own sheets through our models tells you a great deal.
  • Every room with its name and area
  • Wall lengths, door and window counts
  • Text and dimensions read off the sheet
  • An annotated overlay to check it against

What comes back is what our models do before any tuning. It shows you the starting point, not the best we can do — Fine-tuning is what closes the gap.

Send us one sheet

We run it through our floor plan models and send back the extracted data with an annotated overlay, so you can see what comes off your own drawings before anyone books a call.

  • Back in about 2 business days
  • PDF, DXF or a scan
  • NDA on request
This is our base model, before any tuning for you. On a paid project we train it further on your own drawings, which is where the rest of the accuracy comes from. So this shows you the starting point, not the best we can do.

By submitting, you agree to our Privacy Policy. We never share your details.

FAQ

Questions, answered

What teams ask before sending us their first drawings.

More detail in the full FAQ.

  • Not off the shelf. We have base models — models already trained on floor plans, which read walls, rooms, doors, windows, text and symbols — and every project fine-tunes those on the drawings that client actually receives. What you buy is a model tuned to your sheets and delivered into your workflow, which is weeks of work rather than the many months training one from nothing would take.

Get in touch

Tell us what is on your drawings

Describe the sheets you get and what you need off them. We will tell you what our models already handle, what needs fine-tuning on your drawings, and what is not worth automating at all.

Reply time
Within 1 business day

Would rather show than tell? Send one sheet for a free analysis

We reply within 1 business day.

By submitting, you agree to our Privacy Policy. We never share your details.