Diffblue Ltd., the maker of an autonomous generative artificial intelligence code test platform, said today it secured $6.3 million in new funding to build on its differentiated approach to generating unit tests for Java.

The new capital came as an extension to the company’s Series A funding round, joined by existing investors IP Group, Parkwalk Advisors and AlbionVC, with additional participation from Oxford University and further investment from private investors through the Oxford Technology and Innovations EIS Fund.

Citigroup Inc., a customer of Diffblue, joined the round via its venture capital arm Citi Institutional Strategic Investments as a new equity investor. This round brings the total raised to date by the company to more than $46 million.

Founded by researchers from Oxford, Diffblue differentiates itself from other generative AI platforms by using reinforcement learning to automate the error-prone parts of test generation in the software development lifecycle. It works with Java code to produce unit tests that make certain that code will compile and remain bug-free. Because the code it produces is deterministic and it doesn’t use a large language model, but machine learning models, the company says it’s “error-free” and doesn’t suffer from hallucinations.

According to the company, Diffblue Cover, the company’s flagship product, allows customers to automate unit test generation for Java code at once every two seconds, or 250 times faster than a human developer. This means that developers can spend their time on less laborious parts of the development process to ship software faster.

Unit tests are an important software development practice used to verify that individual parts of an application work as expected. These tests are essentially similar to having a manual that explains how to test each part of a prototype car engine design to know that the sparkplugs, distributor, belts, fans and electrical system all work properly before it goes into production. The tests help detect flaws early in development so that they’re easier to fix.

The AI also suggests and automates code fixes designed to improve the understandability of Java code making it more testable. That way future tests will run faster and human developers can maintain it more easily. Since Cover is autonomous, it runs as developers work, meaning that when they change their code, it generates updates and new tests automatically.

“Most of the talk about generative AI in software development is really around a handful of massively capitalized companies pursuing LLM-based technology approaches to solve a broad range of use cases,” Diffblue Chief Executive Toffer Winslow told SiliconANGLE. “Diffblue has chosen a different technical and financial path to address a more focused development problem, albeit one that is the bane of most software teams. The result of our focus is that we are able to solve this problem with a high degree of speed and accuracy without the security and privacy concerns that many companies often have with LLMs.”

For security and privacy, Diffblue said, the code remains entirely on-premises and the machine learning algorithm does not train using it. The operating environment is entirely local, behind the firewall, meaning it cannot leak anything confidential.

“Diffblue is leading the charge in one of the hottest, fastest-moving markets in the history of the software industry,” said Mark Reilly, managing partner at IP Group. “The company’s traction with large, sophisticated customers deploying best-in-class AI technologies across their software development processes is strong evidence of its differentiated, valuable solution.”

In addition to Citi, Diffblue serves 10 of the largest U.S. banks and several of Forbes Global 2000, with key customers that include ING Bank N.V., Cisco Systems Inc., AstraZeneca PLC and the Bank of New York Mellon Corp.

The company said it will launch a new product release for Cover next month that will make it more accessible to developers at companies of all sizes.

Photo: Pixabay