Neosphyr Contact us

About

Computation applied to the built environment.

Neosphyr Ltd is a software company registered in England and Wales, working at the intersection of software, machine learning and the domains where engineering decisions meet data.

The Neosphyr mark

Manifesto

Artificial intelligence is welcome in this industry — on conditions.

Architecture, engineering and construction is a field where decisions are expensive to make, slow to correct and occasionally dangerous to get wrong. That is exactly why it stands to gain from machine learning, and exactly why it cannot accept a system that gives an answer without an account of it. Our position is that the two go together: the usefulness and the accountability are the same problem.

These are the terms we hold ourselves to. They are not aspirations — they are what we are willing to be held to, and what we would expect from anyone selling this industry software that reasons.

A result must show its working

Any conclusion a machine reaches about a building has to be traceable to the values it was reached from. A system that cannot explain itself may be interesting; it is not something to put in front of an engineer who has to sign.

Call it what it is

If software uses a model, say which, on what it was trained, and what it is poor at. If it does not, do not borrow the word. Half of the credibility problem AI has in this industry was created by software that claimed more than it did.

The machine does not carry the responsibility

A person signs the drawing, and no tool transfers that. Good software narrows what a professional has to check and sharpens where they look. It does not stand in for the judgement, and it must never be sold as if it could.

Uncertainty is a result, not a failure

A confident guess is worse than an admission. Where a system cannot establish something it must say so and hand it back to a person — and missing information has to be reported as missing rather than quietly filled in.

Your data stays yours

Models, libraries and project information belong to whoever produced them. We do not train on a client's project data, and we do not treat what passes through our software as material we are entitled to keep.

Defensible a year later

An assessment is often questioned long after it was made. Versions, rules and inputs are recorded so the same question asked again gives the same answer — and so a disagreement can be settled by looking, not by arguing.

Why we are strict about this None of it is a constraint we accept reluctantly. In a field where somebody has to stand behind the result, an answer nobody can check is worth nothing — so an honest system is not the cautious choice, it is the only one that is actually useful.

What we build

Our field is software, artificial intelligence and machine learning applied to two domains that share a problem: architecture, engineering and construction, and financial markets. Both turn on structured data, and in both a decision has to be defensible afterwards.

What we ship today is five Revit plugins, each aimed at the information and coordination work that sits between a model and a decision made on it. They are rule-based and geometric rather than statistical — which is the right tool for the job they do, and we say so rather than attaching the word "AI" to software that does not use it. The products page describes each of them.

Work in financial markets is at an earlier stage and there is nothing to announce yet. When there is, it will be described here with the same precision as everything above.

The thinking behind them

Most of the effort in BIM coordination goes into establishing what a model actually contains, and most of that effort is repeated by hand on every project.

Our tools are built to read the model, report what is there, and put the result somewhere useful — a spreadsheet, a schedule, a parameter — rather than to replace the judgement applied afterwards.

That distinction matters most in BWIC Compliance. The strongest claim the software makes about any opening is that its configuration matches a documented Tested System. That is a statement about the contents of a library, not a statement of legal compliance, and it does not replace professional review.

We would rather say so here than imply otherwise.

The same restraint applies to how we describe ourselves. Machine learning is the direction of the company and the discipline behind how we think about these problems. It is not a label we attach to products that do not use it.

How we work

Report, do not assume

A check that cannot be answered is reported as unanswered. Missing model data is surfaced as missing rather than filled in with a guess.

Your library, your evidence

BWIC Compliance ships with no third-party manufacturer data. The records it assesses against are the ones your own reviewers enter and verify.

Reproducible results

Rule set, library and plugin versions are recorded with every run, so an assessment can be repeated and compared later.

Where your projects already are

Four Revit versions, no cloud account required, no change to how your models are stored or shared.

Company

The details

Registered name
Neosphyr Ltd
Company number
17473213
Registered in
England and Wales
Registered office
71–75 Shelton Street, Covent Garden, London WC2H 9JQ, United Kingdom
VAT
Not currently VAT-registered

Autodesk and Revit are registered trademarks of Autodesk, Inc. Neosphyr Ltd is not affiliated with, endorsed by, or sponsored by Autodesk, Inc.

Want to see one of the tools?

Tell us which product and which Revit version, and we will set up a 14-day trial.