Hiverge, a Cambridge-based startup making it easier for companies to design new algorithms, has raised $5m from Google’s chief scientist Jeff Dean, Ahren Innovation Capital and Flying Fish Ventures, which led the round.
Algorithms are used in everything from data centres and training AI models to designing new chips and mapping GPS systems. But they’re struggling to keep up with the scale of the problems they’re designed to solve, says Hiverge cofounder and CEO Alhussein Fawzi, and to make new ones takes “a significant amount of expertise”.
“If you look at data centres, for example, they’ve been increasing significantly in scale in the last few years, and in order to control them you need algorithms to scale in the same way,” Fawzi says. “We don’t currently have good algorithms to control these ever-scaling systems that can solve problems at millions of times per second.”
Solving the bottleneck on discovery is something Fawzi and Hiverge cofounder and CTO Bernardino Romera-Paredes worked on together at Google DeepMind prior to launching their new company. The company’s CSO and third cofounder, Hamza Fawzi, is a mathematics professor at the University of Cambridge.
The platform that Hiverge has developed, called the Hive, takes code as an input and optimises it for a given output. One example Fawzi gives is training AI models. The user provides the training scripts and the current code, and the Hive then takes that code and tweaks it to make the training run more energy efficient.
“AI training and inference is expensive,” Fawzi says. “This is hopefully going to lead to a massive decrease in the cost of AI by finding more intelligent and better algorithms.
“There’s a lack of people able to do this really advanced algorithmic work.”
Fawzi can’t disclose the specific companies Hiverge is currently working with, but says it’s running pilot tests with clients operating in finance, aerospace and quantum. Potential customers include any large organisation which doesn’t have access to the best algorithmic talent — “this is our sweet spot,” Fawzi says.
As for the business model, he tells Sifted it’s early days and the pricing model might change. It’s currently considering a licence fee model that would see customers pay to run its tech.
The money from the round will go towards developing its product and headcount. The team is currently six, but Fawzi says it plans to recruit mainly across its product team to help bring the tech to market.
It expects to find a lot of that talent locally. “As a former research scientist at DeepMind I noticed that the best research engineers are actually the ones that come from mathematics,” Fawzi says, “so I think there’s a lot of talent we can get from Cambridge University and from the UK more generally.”
