Gesher

An AI that helps engineers write their reports without inventing facts.

It reads a project's documents, answers questions about them, and drafts cited report prose. Every fact traces back to a source; nothing is made up.

Early working prototype, built and tested.
One engineer so far. Looking for a technical co-founder.

Why it works

Harvey did this for law: a frontier AI model plus document retrieval, no custom training. It's now worth about US$11B on ~US$300M revenue, used across the world's largest law firms. The same thing hasn't been built for engineers.

The gap

General AI (Copilot, ChatGPT) invents numbers and citations, useless for a report an engineer has to stamp.
A few startups do one piece, drafting, or document search. The UK's SchemeFlow drafts engineering reports for firms like Stantec and Buro Happold, which shows the demand is real.
No one yet does the whole job, project memory, Q&A, and multi-discipline report drafting, with a hard no-invention rule. That's the opening.

By the numbers

A$18B

Australian engineering & design consulting, the home market (far larger globally).

US$11B

What Harvey is worth doing this for law (~US$300M revenue), the playbook is proven.

75% / 78%

of knowledge workers already use AI at work, most on their own unsanctioned tools (Microsoft, 2024).

~10% / yr

growth of the AEC software market (~US$9 to 12B today).

What's built

The Gesher room: a knowledge rail of cited facts and a drafted, cited paragraph — Drop in a project's documents → it extracts typed facts, each with its source → flags conflicting sources for the engineer (it never guesses) → drafts cited prose → exports to Word, every AI-written passage marked for review.

A real generated geotech paragraph with every value cited to a source standard — A real generated paragraph, every value traced to a source (**AS 1726-2017**, **AS 1289** CBR, **AS 4969-3:2009**). 12 citations; one click to Word.

The plan

Modeled on the playbook Harvey used to scale.

Working prototype, built and tested.you are here
Design partners, one or two real firms running it on live projects, tightening it on their own reports.
First paid pilots → early revenue. Cheap to run; no custom AI training needed yet.
Reinvest & deepen, more disciplines, a custom model, and raise only if it accelerates.

What I'm looking for

One engineer, me, civil, over 13 years in the industry, with the idea, the market research, and a working build. I'm looking for a technical co-founder to build it properly.

It runs single-tenant, inside each client's own environment, so their data never leaves.

adinalieblich@gmail.com

Gesher · co-founder brief · confidential
Sources: Consult Australia / IBISWorld · CNBC / Sacra (Harvey) · Microsoft Work Trend Index 2024 · Mordor / MarketsandMarkets.