Questions, answered properly

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.

Straight out of the box on a typical plan our base models are generally above 80% on the elements they were built for, which is enough to tell whether a job is feasible but not enough to run unattended. Fine-tuning on your drawings is what closes the gap, and we agree a target on the specific fields you care about before starting rather than quoting a headline number.

Vector PDFs, DXF exports, and scanned or photographed plans. Where a DXF is available we read the vector geometry directly, which gives exact lengths and areas. For PDFs and scans we combine segmentation, object detection and OCR of the printed dimensions, and calibrate against the scale note on the sheet.

Structured JSON or CSV with rooms, walls, doors, windows, text and symbols, plus an annotated overlay so you can check the result against the drawing. Every value keeps a link to where on the sheet it was found, so a disagreement can be settled by looking rather than re-measuring.

General vision-language models handle the spatial layout of a high-resolution drawing poorly and tend to invent counts and dimensions rather than admit uncertainty. We split the problem instead: segmentation for walls and rooms, detectors for symbols, OCR built for rotated drawing text, and a language model only on top of the structured result.

Yes. We sign an NDA with every client, and the pipeline can run as an API in your own cloud account or on your own servers, so drawings never leave your network.

Send one sheet through the form and we run it through our base models, before any tuning for you. You get back the extracted data and an annotated overlay, usually within about two business days, with no call required. It shows you the starting point on your own drawings, which is a far better basis for a decision than an accuracy figure measured on somebody else's plan.

A proof of concept on your own plans typically takes four to six weeks. We agree the fields and the accuracy target up front and measure the result against your manual work on a held-out set of your sheets.

A representative sample of the drawings you receive, including the awkward ones, and a description of the fields you want. If you have existing manual results for some of those sheets they are valuable, because they become the ground truth we measure against. We handle the labelling.

You own the weights we train for you and the data you give us. We do not reuse a client's drawings to train models for anyone else.

Yes, and it is one of the more useful things to automate. Door and window tags detected on the plan views are matched against the tabular schedule sheets, which surfaces tags that appear on one and not the other. A reviewer confirms the discrepancies rather than the system acting on them.

Yes. Sheets are identified from the title block, plan views are separated from schedules and details, and results are consolidated across the set. See construction drawing analysis for how the sheet-level pieces fit together.

Yes. Two issues of the same sheet are aligned and compared element by element, so what changed is reported as a list of added, removed and moved elements rather than a visual diff a person has to interpret.

Per project, based on the fields you need and how varied your drawings are. There is no per-page price list, because a pipeline for one residential plan type and a pipeline for mixed commercial sets are not the same job. The free analysis and the scoping conversation both cost nothing.

Floor Plan Analyzer is the floor plan practice of AxcelerateAI, an AI engineering company with offices in Austin and Lahore. The same team builds the computer vision, OCR and deployment work across the parent company's other domains.