Exclusive: Ex-Google DeepMinders’ algorithm-making AI company gets $5 million in seed funding
Two former Google Deepmind researchers who worked in the predhafal -winning Alphafold Predictions of the Nobel Prize Prize Company, as well as the Alphavoville Code, with a new company, with the task of accessing advanced algorithms.
The company, called Hiverge, originated from STELTH today with $ 5 million in seeds financing, led by Fish Fish Ventures with the participation of AHREN Innovation Capital and Alpha Intelligence Capital. The legendary programmer and the chief scientist of Google Jeff Dean is also an investor in startup.
The company has built a platform called “HIVE” that is used AI to create and test new algorithms to operate vital commercial operations – everything from the product recommendations to delivery – automatically improving it. While large companies that can use their data science and machinery teams sometimes develop detailed algorithms, this ability was far from the reach of most medium and small companies. Smaller companies often have to rely on ready -made programs that come with pre -designed algorithms that may not be perfectly suitable for this particular work and their data.
The HIVE system is also the possibility of discovering unusual algorithms that may produce superior results that human scientists may not be able to develop through intuition, experimental and error, said Alhussein Fawzi, a founder and executive manager of the company. luck. He said: “The idea behind Heferg is to enable these companies with the best best algorithms in their class.”
“You can apply [the Hive] “To the algorithms of machine learning, and then you can apply them to the planning algorithms,” Fawzi explained. They are the two things, in terms of algorithms, completely different, however, they actually improve each of them. “
In Google DeepMind, Fawzi led the team that developed in 2022 alphatensor AI, which discovered new ways to hit the matrix, a basic sporting process for training and operation of nervous networks and many other computer applications. The following year, Fawzi and the team developed Funsearch, a method that used large language models to create new coding methods and then used an automatic evaluation to get rid of wrong solutions.
It has also worked on the early stages of the Google DeepMind, which uses many LLMS that works together as agents to create a complete new symbol rules to solve complex problems. Google did the idea of alphavolve while finding ways to improve LLMS. For example, I found a way to improve the method of Gemini to hit the matrix to provide 23 % speeding up; He also improved another major step in the transformer, which is the type of artificial intelligence structure on LLMS, work, and reinforcing speeds by 32 %.
His brother Hamza Fawzi, a professor of applied mathematics at Cambridge University, who works as a technical advisor for the company; Bernardo Romera-Predes, which was part of the Google DeepMind team that created Alphafold and who is now the chief technology official in Hiverge.
Hiverge has already shown the benefit of its HIVE system by using it to win the Airbus Beluga challenge, which calls on the contestants to find the best way to download and store parts of the aircraft carried by the Airbus Beluga XL. The solution was developed by HIVERGE to accelerate 10,000 times on the current aircraft loading algorithm. The company also showed that it might take the algorithm algorithm already improved and accelerated another three times. I have found new ways to improve computer vision algorithms.
Alhussein Fawzi said that Hiverge, based in Cambridge, England, currently has six employees, but it will use the money collected in the last financing round to expand its team. “We will also move from research to building our products,” he said.
The company plans to make the technology available through cloud markets such as AWS and Google Cloud, where customers can use the system directly on their code. The statute analyzes the parts of the code that represent bottlenecks, generate improved algorithms, and provide recommendations for engineers.
2025-09-16 23:01:00
