Leo AI

There is a familiar pattern in the current AI boom. Someone takes a large language model, points it at a profession, and declares it solved. The trouble is that a model trained on the internet is very good at sounding right and only occasionally good at being right. For a marketing email, the gap is survivable. For a load-bearing bracket, it is not.

Leo AI's founders noticed this gap and did something unusual about it. Rather than wrap a general chatbot in an engineering costume, they set out to build what they call a Large Mechanical Model - an LMM, if you enjoy acronyms that rhyme with the thing they are replacing. Instead of learning from books and websites, it learned from millions of real-world products. Its vocabulary is made of bolts, bearings, brackets and the standards that govern them.

The result, the company says, is a co-pilot that behaves like an engineer rather than a very confident intern. It integrates directly into CAD software and PDM systems - the vaults where a company's design history quietly accumulates - and turns text, sketches and specifications into production-ready 3D models. It answers technical questions, runs mechanical calculations, sources parts by plain-language description and drafts design concepts.

The accuracy claim is the part that makes people lean in. Leo says it answers engineering questions correctly about 96% of the time, against roughly 46% for generic AI tools. Numbers like these are always worth a raised eyebrow - benchmarks are chosen by the people they flatter - but the direction is the point. A model that learned mechanical engineering should be better at mechanical engineering than a model that learned everything. This is not a controversial idea. It is just an expensive one.

What makes the story more than a spec sheet is where it came from. The company's co-founders are mechanical engineers who met in an elite Israeli military technology program. CEO Maor Farid holds a PhD in mechanical engineering, did postdoctoral research at MIT as a Fulbright fellow, and - a fact that keeps appearing in every profile - was the youngest PhD graduate in the history of the Technion. Before Leo, he was an engineer digging through old folders and vendor catalogs, manually reusing designs, hunting for the right standard. Leo is, in a very direct sense, the tool he wished he had.

That origin explains the company's posture, which is refreshingly un-apocalyptic. Leo's stated philosophy is that "AI won't replace mechanical engineers, but AI-empowered mechanical engineers will replace those who aren't." This is a tidy piece of positioning, but it also happens to be a reasonable reading of how these tools actually land in a working engineering department. The tedium goes; the judgment stays.

Investors have found the argument persuasive. In September 2025 the company closed a $9.7 million seed round led by Flint Capital, with participation from TechAviv, Two Lanterns VC, OurCrowd and Mento VC, plus a roster of strategic angels. The most telling name on that list is Bertrand Sicot, the former CEO of SolidWorks - which is to say, a person who ran one of the incumbents Leo is politely trying to disrupt. His verdict: "Leo is the first AI applied to engineering, and engineers are dying for this progress." When the old guard writes checks to the new one, it usually means something.

The traction supports the enthusiasm. Leo says more than 50,000 engineers now use the platform, and that it generated over $100,000 in its very first month of monetization without spending on paid marketing - the kind of pull that suggests real demand rather than manufactured hype. Its customer list reads like a tour of things that move: Scania, HP, Mobileye, and, per some accounts, Toyota and Philips. In 2026 it was ranked the #1 new AI software globally on G2.

None of this means Leo has won. It is competing against generic assistants that are free and getting better, and against CAD incumbents - Autodesk, Dassault, PTC, Siemens - who own the software engineers already live inside and are busily bolting AI onto it. The bet Leo is making is that depth beats breadth in a domain where being wrong has consequences you can measure in newtons. It is a good bet. It is not a sure one. But of all the places to point a model that actually understands its subject, "the physical world" is a pretty compelling choice.