Largely, yes, and the honest version of that answer is more useful than the headline. An AI civil engineer can produce the whole noise assessment end to end, not just help with a step of it: it reads the material, runs the standard methods, checks its own numbers, and hands back a complete draft. What it does, what it does not, and what the engineer still signs, below.
The deliverable
A noise assessment answers a planning question: what sound levels a new development will be exposed to, or a new road or railway will cause, and whether they stay within the guideline values that decide if it can be built. It compiles the sources and the site, models how the sound propagates, computes the equivalent and maximum levels at each facade and outdoor space, checks them against the limits in the traffic-noise ordinance, and recommends mitigation where they are exceeded. It follows documented methods and standard models, gets checked by a colleague, and carries the signature of an engineer accountable for it. Most municipalities require one before approving a development near a road or railway.
The work is less exotic than the word suggests. It is many kinds of work in sequence: gathering the underlag, building the model, running it again each time the plan moves, compiling the document, checking every figure against the right guideline. All of it is engineering, and none of it is invention. The craft is applying a known method to one specific place, carefully, across many steps, and that is exactly what makes the whole thing a candidate to hand over.
What the AI does today
It produces the whole draft. Handed the assignment, an AI civil engineer reads the project material and the governing documents, pulls the traffic figures and source levels from where they live, runs the standard propagation calculation, compares the levels against the guideline values, drafts the report into the client's template, and checks the output against its own numbers before anyone sees it. Not a fragment sped up: the deliverable, produced end to end and handed back for review.
That is the line between this and the tools that came before. A copilot answers a question or speeds one calculation, and you still gather the inputs, move results between tools, and assemble the document yourself. Here the unit is the whole task. The engineer briefs it and gets back a draft, the way they would from a capable colleague, in an afternoon rather than across weeks.
What it does not do matters just as much. It does not decide whether the method suits the site. It does not weigh a mitigation against what the municipality will accept, or what a local engineer knows about that setting. It produces the written deliverable and works from the model and the map data; it does not decide the barrier. And it is not finished when the status line stops: the draft lands with an engineer for review, the way a colleague's would.
What stays human
The judgment that makes the assessment worth signing. Whether the chosen method fits this site. Whether the results hold together. What mitigation to actually recommend, and how to defend it through a planning process. That judgment was always a small share of the hours and the whole reason for the deliverable, and it does not transfer to the machine.
The signature stays human: an engineer reads the draft, questions a figure, sends corrections back, and puts a name on it. What changes is the review around that signature. Human checking runs one pair of eyes at a time, and reviewers tire; a system that re-derives every calculation independently does not, so it flags deviations a tired pass can miss. The catch-rate rises; the accountability stays put.
There is a quieter reason the engineer cannot be removed. Yesper learns the method from public codes and standard practice, and it learns from the verdict: what the accountable engineer corrected and accepted before signing. It does not learn from your project data. The engineer is not only the reviewer; they are the reason the next assessment is better.
The economics
Not the scrutiny, and not the deliberation. What collapses is the price of a second try. Noise assessments are usually sold at a fixed fee against an hour budget, which makes the revision loop the most expensive thing in the project. The plan moves: a building grows a storey, a facade turns, a barrier is added, and a week of modelling has to be redone by hand. Those absorbed re-runs, not the engineering, are where fixed-fee assessments quietly bleed.
When re-running the whole calculation chain costs an afternoon instead of a week, two things change. Revision loops stop being toxic, so a moving plan is no longer a threat to the fee. And an alternative that used to be dropped because evaluating it cost too much now gets evaluated: one more barrier height, one more building layout. Same fee, one more iteration, a better-tested recommendation. The waiting goes; the judgment stays.
FAQ
Can AI write a noise assessment on its own, without an engineer? No. It produces a complete draft and checks it against its own numbers, but the assessment is signed by an accountable engineer who reads it, questions it, and owns the result. The point is an engineer freed from producing the document by hand, not an unsupervised machine.
Is it accurate? It runs the same standard models an engineer would, then self-checks the document, re-deriving each figure rather than trusting it. Accuracy is confirmed by the signing engineer; what is new is a relentless second reviewer running in parallel with the human one.
Do you train the model on our project data? No. Yesper learns the method from public codes and standard practice, and it learns from the verdict: what your engineers correct and accept before signing. Your project data is the compass, never the fuel.
Can it write other deliverables, not just noise assessments? Yes. Several deliverable types are already in production; the noise assessment is just the one this page goes deep on. We won't post a page for every type, but the pattern holds across them: the whole document is produced end to end, and an engineer reviews and signs it.
How long does it take? The draft comes back in an afternoon of machine time rather than the weeks a hand-built assessment can take. The engineer's review is separate, and it stays as thorough as the project needs.
Yesper is the AI civil engineer for construction and infrastructure. On a noise assessment, today, that means the draft comes back in an afternoon, and the engineer's hours go where they matter most: the judgment, the alternative worth weighing, the recommendation worth standing behind.
The article describes what Yesper actually does with a noise study, and where the engineer takes over. If you'd like to see it in a demo, get in touch.
Book demoNot ready for a demo? Get the next piece in your inbox.
