JuliaHub Inc., an “agentic” industrial engineering startup that’s trying to automate complex manufacturing processes with artificial intelligence, said today it has raised $65 million in a Series B funding.

Today’s round was led by Dorilton Capital and saw the participation of General Catalyst, AE Ventures and the technology industry luminary Bob Muglia, the former chief executive of Snowflake Inc. The funds will advance the development of Dyad 3.0, an AI agent platform that’s meant to help engineers design, test and maintain complex hardware such as semiconductors, satellites, lithium batteries and more.

Though software engineers have been given a massive helping hand by AI coding bots such as Claude Code and GitHub Copilot, their counterparts in the world of industrial machinery – the people designing and building aircraft engines, heat pumps, wastewater facilities and so on – have been left behind. They’re still using legacy design tools and simulation platforms, and their design timelines often stretch into months or even years.

The result is that the world is facing a massive “infrastructure gap” that will require a cumulative investment of more than $106 trillion by 2040, according to a study by McKinsey & Co. last year. But JuliaHub argues that those engineers will need more than just money – they’ll also require new design tools that allow them to move at the same breakneck speed of software development. That’s where Dyad comes in.

Agentic engineering at scale

Dyad 3.0 is designed to enable what JuliaHub Chief Executive Viral Shah calls “agentic engineering at scale.” It’s not just a chatbot, but rather a cloud-based environment that provides a home for vast numbers of AI agents to get to work in designing the world’s industrial infrastructure. It’s grounded in the laws of physics and can be used to create ultra-realistic systems and lifelike environments for stress testing new machines and infrastructure.

In tests, Dyad was able to automate the entire design process of a new system of model-predictive controllers used in chemical manufacturing plants. That would normally take months of manually intensive work if humans were doing it.

The startup’s secret weapon is the Julia programming language, which was designed for high-performance mathematical computing. According to Shah, it’s what allows Dyad to combine scientific machine learning with scalable physics simulations, so they can feed a full specification of what they want to build into the system and let the AI agents do all of the design work.

“It’s not about helping engineers complete one small task at a time,” Shah said. “It’s agentic engineering at scale, where teams can feed a full specification to Dyad and have it design the complete system. Spec in. Design out.”

One of JuliaHub’s biggest challenges is the problem of AI hallucinations. Though an AI assistant can be forgiven for making syntax errors when it’s drafting a blog post or PowerPoint presentation, the same kind of leniency is not possible in the real world. If an agent involved in the design of a new bridge makes an error, it could cause the structure to collapse and kill dozens of people, for instance. To get around this, the agents must have a complete understanding of advanced concepts such as gravity, thermodynamics and fluid dynamics.

Scientific machine learning is the company’s solution to this challenge. It refers to a hybrid approach that blends data from real-world sensors with physics-based equations to ensure that model’s outputs are accurate, even when conditions change and new factors are entered into the mix.

The Dyad platform provides agents with access to complex scientific tools and vast amounts of data. That enables them to develop “digital twins” of systems and then automate the stress tests required to ensure they can stand up to everything the real world throws at them.

Early promise

The startup says Dyad has already yielded some impressive results. Working with the water management company Binnies, JuliaHub built a digital twin of a complex water pump system that can predict failures with more than 90% accuracy, despite having only four sensor inputs to glean data from.

JuliaHub has also worked with the semiconductor design software firm Synopsys Inc. to enhance chip development. Synopsys Senior Vice President of Innovation Prith Banerjee said Dyad has helped to transform system-level engineering. “It enables high-fidelity hybrid digital twins by integrating physics-based simulation with data-driven models,” he said. “What once required extensive manual effort can now be done far more efficiently, accelerating the entire digital engineering lifecycle.”

The startup said its goal is to make Dyad the industry standard for “AI-native engineering” and to do that, it will use the money from today’s round to scale its go-to-market efforts and enhance its integrations with partners. In future, it aspires to go beyond just designing the world’s infrastructure and enable “autonomous operations.” It imagines a world where every complex machine is managed by AI agents that can optimize its performance and anticipate problems, with only minimal human intervention.

Images: JuliaHub