SHEET 10AboutAI Agentiva · Dubai, UAE

Built around the estimator.

One question started this: can software help an ELV estimator spend less time counting and more time deciding?

Not “can a model read a drawing” — that part is interesting but solved enough. The question is whether the output can be trusted by someone whose name goes on the tender, and what a tool has to give up to earn that.

AThe decisions, and what they cost

Every answer here has a trade-off attached.

A “why” with no cost in it is marketing. These are the five decisions that shaped the product, and the thing each one made harder.

01Why ELV: It is dense, repetitive and rule-governed enough to model honestly — a door really does imply five devices, outlets really do aggregate into patch panels — and it is underserved by software built for structural or MEP. ELV takeoff is where the reasoning is regular enough to encode and the manual effort is high enough to be worth removing.
Cost
It also means starting narrow, in one discipline, instead of claiming to do every trade badly.
02Why drawings: Because that is where a tender starts, and where the quantity exists before it exists anywhere else. Schedules and specifications describe the drawing; they do not replace it. Anything that begins downstream of the sheet inherits somebody else's count and cannot check it.
Cost
Reading drawings properly — scale, legend, layers, revisions — is much harder than parsing a spreadsheet. We took the harder input on purpose.
03Why human verification: Because responsibility cannot be delegated to a model. In a tender review the question is not whether a number is probably right, it is who stands behind it. A machine observation is a claim; only a person can turn it into something they are willing to defend.
Cost
It keeps a human step in a pipeline a competitor can advertise as fully automatic. We think that step is the product.
04Why deterministic rules: Because auditability requires reproducibility. Run the same versioned rules over the same verified observations and you get the same quantities — today, and in eight months when a variation claim turns on how something was counted. A second model in that position would give a plausible answer and no account of itself.
Cost
Rules must be written, maintained and argued about. That is slower than asking a model, and it is the only way the output survives scrutiny.
05Why the GCC: Because this is where we are, because tender cycles here are fast and unforgiving, and because a consultant in Dubai has a fair question about where their client's drawings are being processed. We would rather answer that with in-region infrastructure than with a paragraph in a privacy policy.
Cost
Building in-region is more work than renting a US endpoint. See Security & data for exactly what is implemented today and what is not yet claimed.

BFounders’ notes

Zakaria

Co-founder · COO

Builds the product

I build the part of this that decides what a number means, so I am the one who has to defend the step everyone wants us to remove.

The tempting version is a straight line: upload a drawing, detect symbols, total them, export a BOQ. It demos beautifully. It is also the version I will not ship — because the first time an estimator is asked in a tender review where a quantity came from and the honest answer is “the model said so”, the tool is finished, and so is the person who used it.

So the system is built the long way round. Detection proposes; it never concludes. A person confirms, corrects, reclassifies or adds, and every one of those actions is attributed and timed. Quantities come out of rules that are versioned and readable, not out of a second model, because determinism is the property that makes a number auditable rather than merely plausible.

That costs us a step our competitors can skip in a demo. It is also the step the whole thing is worth.

Mohammed Abdul Azeez

Co-founder · CEO

Company and customers

Every tender in this region starts the same way. A package lands, the deadline is short, and somebody has to count.

Not count in the easy sense. A camera is a symbol on a plan, a model number in a specification and a row in a schedule — and those three can disagree with each other. A door is one line from the architect and five devices in the ELV package. The disabled toilet alarm quantity is not on the ELV drawing at all; it is derived from how many accessible WCs somebody else drew. That is skilled work, done under time pressure, by people who are simultaneously expected to be judging risk and pricing strategy.

What they are not short of is opinions about AI. What they are short of is hours, and a way to show their working when a number is challenged.

So the ambition here is narrow and specific. Not to replace the estimator — to give back the hours currently spent counting, so they are spent deciding instead.

CThe company

Small, technical, and in the same time zone as the tender.

Company: AI Agentiva
Based: Dubai, United Arab Emirates
Focus: Engineering quantity take-off
First discipline: ELV
Founders: Two
Stage: Building with early customers

The wedge is ELV, because that is where the reasoning is regular enough to model properly and the manual effort is worth removing. The direction is a quantity takeoff platform that can extend into other disciplines — but only once each one is understood as well as this one, which is a slower promise than the usual.

Bring a drawing Talk to us