Leo AI: How CAD-Aware AI is Changing Mechanical Design and Engineering Workflows

A scheduled Zoom call with Maor Farid, Ph.D., co-founder and CEO of Cambridge-based Leo AI, began with an email pitch. Farid had read an article on Machine Design, and noted that while he enjoyed reading the piece, he believed “AI’s impact on product design over the next two years to be more nuanced than the piece suggests.”

Machine Design took the bait. Exploring the evolution of CAD design and the influence of technologies such as AI on mechanical engineering is table stakes for our editorial team.

So, what can Leo AI offer design engineers? What problem does it solve?

“When we talk to engineers, and as engineers ourselves, we see that those generic tools are just not enough,” said Farid. In other words, tools such as ChatGPT and Claude are not built for the complexity of mechanical design workflows.

Leo AI, a patented Large Mechanical Model (LMM), is a “CAD-aware AI trained on trained on customer-specific models and documentation, plus more than 1 million vetted engineering sources. Leo AI achieves 96% accuracy (compared to GPT’s 46%) and saves engineers an average of more than five hours per week, Farid touted.

Backed by three patents and branded as a Large Mechanical Model (LMM), Leo AI was built with the aim of transforming “the entire product design process into an AI-centric one, where Leo serves as a natural extension of the engineer’s mind,” noted Farid.

Since launching Leo AI in May 2023, the startup based in Cambridge, Mass. has grown to more than 60,000 users and raised more than $9.7 million in seed funding, delivering more than 340 updates annually. Their client base includes companies such as Scania, HP and Mobileye.

Farid, together with his partner Moti Moravia, Leo AI’s co-founder and CTO, shifted from traditional mechanical engineering careers to roles centered on the intersection of AI and mechanical engineering.

“We’re building an engineering-grade AI specifically for engineering leaders and firms,” he said. Leo integrates PLM systems and understands CAD (including text, images and actual CAD representations), which allows engineers to access legacy information and avoid redundant work, explained Farid.

“The problem is not that SOLIDWORKS is not great, not that Onshape is not great, but the fact that in 2025, our knowledge—the tremendous amount of knowledge in the organizational databases—is huge, but it’s still siloed,” said Farid.

Engineering knowledge is often locked in the minds of experienced mechanical engineers and legacy documents their organizations created over the course of decades, but they’re also contained in “the mountains of CAD files that they’ve designed,” he pointed out.

READ MORE: AI Enhancements to CAD and CAE: Faster and Better Design with More Possibilities

“If you want to see the minimal thickness of the hull of the battleship that you designed in 1995 was, it’s not written in any PDF document,” said Farid. “You need to open Onshape or SOLIDWORKS and measure it.”

Or consider the time an experienced mechanical engineer spends trying to add a single spring to a suspension system. The part may already exist in the company’s PLM system, said Farid, but the engineer is limited in the ways he can efficiently access organizational knowledge, vendor catalogs and PLM data.

Farid estimates that inefficiencies in accessing organizational knowledge waste about 40% of engineering time, cause 40% of time-to-market delays and add 25% to product lifecycle costs.

CAD-Aware AI that Understands Geometry and BREP

Leo AI processes not just text or images but also the BREP (boundary representation), which is detailed geometrical and topological representations of the CAD parts and assemblies. AI platforms like ChatGPT tend to hallucinate and rely on sources such as Reddit, argued Farid.

Since Leo AI is purpose-built for mechanical engineering workflows, it mitigates these issues by streamlining part sourcing and decision-making. “It was trained on over 1 million bibles of engineering and 120 million vendor parts, and it’s connected into the organization’s knowledge bases, their PLM, Windows directories, etc.,” said Farid. “And in this way, it basically can answer any question on your data, answer technical questions and find you the best-fit parts and everything in enterprise-grade security.”

Security Check: Compliance with SOC 2 and GDPR

Leo AI conforms to enterprise-grade security standards, including SOC 2 and GDPR compliance. When connected to PLM systems, document libraries and directories, Leo AI does not move or duplicate files. Instead, it uses embedding vectors to index CAD and documents. The LMM treats machine parts as tokens (mechanical components such as bolts, bearings, spindles, etc.) and uses them to construct an assembly model. Moreover, said Farid, these numerical representations cannot be reverse-engineered, so customer data remains private. 

Assembly Inspector Function

During the demonstration, Farid showed how the copilot enables an engineer to review designs before submitting them for approval. By clicking Leo Inspect, the tool analyzes the design and evaluates how well each part aligns with organizational best practices and inventory standards. Engineers can view detailed information about suggested parts, including drawings and previews and seamlessly replace components with a single click.

Farid showed the path to finding a spring for the design of a suspension system that needs an 8mm bushing. The engineer prompts with requirements and the copilot automatically lists parameters and formulas, along with references.

READ MORE: Generative AI Catalyzes Growth in Industrial Robotics & CNC Machining

Leo AI calculates the parameters and finds the optimal part from the PLM, which is already vetted inside the organization, and suggests alternatives that better match geometry, requirements and company standards. “It also shows vendor parts in order to give you the full perspective,” added Farid. “Then, it finds you the relevant parts that fit your best practices and your current CAD design.”

If, for example, one works with a bronze bushing and needed to determe the correct force, taking into account factors such as heating or cooling, tolerance and material properties, Leo AI steps in by outlining its approach and provides a definitive answer backed by credible sources. Should the user click on a reference, it opens to the associated page in that source.

What makes this powerful, explained Farid, is that every recommendation from Leo is backed by authoritative sources. “Leo might reference a calculation from Joe’s 1999 report” or cite an internal standard to justify the required force, he said.

Integration with CAD and PLM Systems

Leo AI was designed to plug directly into the CAD environment. The company recenty introduced Onshape integration, which will be available soon, said Farid. “There are open-source CAD integrations available online,” he noted. “For example, the integration to SOLIDWORKS is available publicly available online.”

Beyond off-the-shelf options, Farid’s team builds custom integrations for enterprise customers whose PLM and PDM systems and internal document libraries are tailored to their operations. Each deployment is adapted to the organization’s requirements and Leo AI handles the rest of the integration.

For design engineers, said Farid, this means a reduced workflow friction; fewer manual file-handling errors; and access to parts, models and documentation without leaving the design workspace.

End Goal: Automate the “Boring 80%” of Engineering Work

Engineering is built on questioning and validating every decision, Farid said, so full automation is not the goal. Farid has a clear vision for what Leo AI should ultimately become. Leo AI is meant to handle the “boring 80%” of engineering work such as sourcing parts, vetting standards and running calculations. This frees up time so human engineers can focus on more creative endeavors, such as conceptualization, innovation and orchestration.

READ MORE: ML-Enhanced Computer-Aided Engineering in the Cloud

Even in the long term, he said, Leo will be programmed to generate small, inspectable assemblies that enable engineers to vet and iterate with just a handful of questions or prompts.