Frequently Asked Questions

ArchLab's deterministic extraction pipeline achieves sub-centimetre accuracy when working with calibrated inputs (photographs with a known reference dimension or properly scaled scans). For uncalibrated smartphone photographs, accuracy depends on image quality and angle, but proportional relationships between dimensions remain consistent. Importantly, when our system cannot determine a measurement with high confidence, it flags the uncertainty rather than guessing. This means you will never receive a fabricated dimension — you will receive either a confident measurement or a clearly marked uncertainty flag that prompts manual verification.

ArchLab accepts a wide range of input formats: JPEG, PNG, WebP, TIFF, HEIC (iPhone photos), and PDF. This includes hand-drawn sketches on any medium (napkin, whiteboard, graph paper), scanned architectural drawings, smartphone photographs of rooms and spaces, tablet drawings, and even low-quality copies of existing plans. Our intake engine normalises orientation, corrects lens distortion from photographs, and handles varying image quality levels.

ArchLab exports to DXF and DWG formats, which are natively compatible with Revit, AutoCAD, ArchiCAD, and virtually every other CAD and BIM platform. Output files are layer-separated (walls, doors, windows, dimensions, annotations on separate layers) following standard architectural drawing conventions. We are also developing direct Revit family integration and IFC export for BIM workflows.

Data security is a core priority. All data is encrypted in transit (TLS 1.3) and at rest (AES-256). Project files are processed in isolated environments and are not used to train our models. We do not share project data with third parties. For firms with strict data governance requirements, we offer an on-premises deployment option where your data never leaves your network. We are on a SOC 2 Type II compliance roadmap with expected certification in 2027.

Generative AI tools create images by predicting what something should look like. They produce beautiful visuals but fabricate dimensions, proportions, and spatial relationships. A room rendered by Midjourney might look photorealistic but be three metres wider than reality, have doors that open into walls, or windows that float above the floor line.

ArchLab uses deterministic computer vision — not generative AI — to extract real measurements from your inputs. Every dimension is derived from geometric constraint solving. When we do generate renders (for visualisation use cases), they are built on top of the verified spatial data, so proportions and dimensions are accurate. The fundamental difference: generative AI imagines spaces; ArchLab measures them.

We are currently in pilot phase and have not announced public pricing. Waitlist members will receive founding-member pricing that will be locked in for the lifetime of their subscription. For firms interested in piloting ArchLab, we offer a free evaluation on one of your real projects — we process your files and deliver the output so you can evaluate quality before committing. Contact us through the firm demo form for details.

Yes. ArchLab can process inputs floor-by-floor and generate separate floor plans for each level. Our system maintains spatial consistency between floors (stairwell locations, column alignments, structural grids) when processing a multi-storey project. For our C3D pilot project, we successfully processed a three-floor commercial building from smartphone photographs taken during a single 90-minute site visit.