While everyone talks about DeepSeek and its rival OpenAI, what both don’t have (yet) is the ability to tackle niches. These are general platforms that are great for productivity but aren’t quite fine-tuned to tackle industries like construction – more specifically structural design.
A Los Angeles based startup, Genia, is launching “Genia Structural CoPilot,” a generative AI model targeting the construction industry. While it’s not for broader use like ChatGPT, it’s still highly important in its own way. For instance, the construction industry is coming to grips with increasing natural disasters – take the Palisade fires. Events like these surprisingly overwhelm firms who suddenly get inundated with requests for new constructions. Behind the scenes however, the progression from coming up with an architectural design, all the way to getting approvals from local regulators is just the first of a long process that often takes months or sometimes years.
Traditionally after designs are drafted by architecture firms, they’re sent to structural engineering consultancies. This is where concepts must be converted into a structurally sound design. And this process is not only time consuming, but also labor-intensive. It could take 15 engineers roughly 18 months to design a 20-story tower. That could mean 750,000 manual calculations. On top of this, the designs could be rejected by local regulators as frequently as 40% of the time. And more often than not, this is due to human error in the calculations and design.
With $3 million in funding led by Pi Labs and supported by Amplify, Boost VC, Dorm Room Fund, Suffolk Technologies, Y Startup Index, Moatable founder Joseph Chen, and Scale AI, Genia aims to radically speed the structural design process up by 10 times thanks to the development of its own generative AI model. Simply uploading AutoCAD or PDF files from architecture firms, Genia Structural CoPilot is capable of rapidly outputting up to five variations of physics-validated structural designs. The variations come from optimizations around cost, sustainability, among other criteria. And each comes with a suggested list of detailed construction materials and optimized so that you’re using 20% material than through manual calculations.
“Mounting cost pressures on construction projects, and the need to deliver housing across the world, mean that engineers and developers must increase both the quality and volume of output in a commercially viable manner,” said Faisal Butt, Managing Partner and Founder of Pi Labs. “Generative design is critical to solving this bottleneck and we believe Genia’s solution will become the go-to tool for thousands of companies in major development markets. It could be a game changer in facilitating offsite construction, enabling shorter build times, lower costs and more sustainable carbon profiles for projects.”
What’s groundbreaking about Genia is that its generative AI model was built from the ground up by its team, co-founded by ex-Amazon engineer, Zhihao Zhao, and ex-Arup structural engineer Robin Li. Other team members include software entrepreneur Houtao Wang and generative AI researcher Peter Dai. The AI is trained on a combination of data points including designs that are open source and from clients. But the company also has an in-house team that creates its own designs purely to bolster its AI’s training. The data is then validated by cross referencing it through an industry-first partnership with Weyerhaeuser's ForteWeb and Simpson Strong-Tie, which are among the largest building material suppliers in the U.S.
Talking about general AI models like GPT4, Llama or others, Zhihao Zhao, Co-Founder and CEO of Genia explains, “they’ll tell you that you need to have a beam to connect the walls, but it can’t tell you what type of beam and the exact measurements and coordinates that the beam needs to be designed with. This is where Genia comes in with pixel-level accuracy and offers physics-backed measurements.”