Hiverge, a Cambridge-based startup founded by former Google DeepMind researchers, has raised $5 million in seed funding to develop AI algorithms that optimize backend code and address scaling challenges in AI and data systems.

The funding round was led by Flying Fish Ventures and included notable investors like Jeff Dean from Google, as well as Ahren Innovation Capital, with the investment completed swiftly within two to three months.

Hiverge's technology is designed to make AI training and inference more cost-effective by discovering smarter algorithms, helping industries like finance, aerospace, and quantum computing overcome bottlenecks caused by scarce algorithmic talent.

With the recent funding, Hiverge aims to accelerate market deployment, expand research efforts, and develop new products, inviting businesses to join its early-access program for innovative algorithm solutions.

Founded in late 2024 and emerging from stealth, Hiverge is still experimenting with its business model, primarily planning to license its platform to customers.

The company was founded by experts including CEO Alhussein Fawzi, CTO Bernardino Romera-Paredes, and Chief Science Officer Hamza Fawzi, all with backgrounds at DeepMind and Cambridge University, aiming to automate algorithm design to surpass human capabilities.

The company’s platform, called 'The Hive,' uses program synthesis to automatically generate, test, and optimize algorithms, significantly improving AI training efficiency—by over 15% in some cases—and enabling faster resource planning, as demonstrated in a project with Airbus.

Hiverge's founders, with a background in pioneering projects such as AlphaTensor and AlphaEvolve, aim to automate and improve algorithm discovery, leveraging their expertise to create more efficient AI solutions.

Hiverge plans to expand its team and transition from research to product development, making its technology accessible via cloud marketplaces like AWS and Google Cloud to help businesses optimize their code.

The startup is currently testing its technology with clients in sectors such as finance, aerospace, and quantum computing, targeting large organizations that lack advanced algorithmic expertise.

Hiverge emphasizes assisting engineers with backend code improvements rather than creating user-facing applications, focusing on verifiable, efficient code rather than vibes-based AI.

The founders have a strong history in AI and algorithm discovery, contributing to projects like AlphaTensor and FunSearch, which have improved efficiencies in data centers and chip design.

Hiverge’s platform leverages program synthesis to generate and optimize algorithms, addressing limitations of current solutions that are often expensive and require scarce expertise, thereby enabling better decision-making across industries.