For any company using software, there is often a difficult balance between patching systems as quickly as possible to prevent cyberattacks and making sure those updates do not break how the systems work. This is where the startup CyDeploy wants to help.
The company’s founder, Tina Williams-Koroma, explained that the idea of CyDeploy is to use machine learning to observe and record the most important systems a company runs. It creates a replica of them, allowing the company to run tests on what she called a digital twin before deploying updates on the real live systems.
She stated that they record how users are using applications and systems on a regular day-to-day basis. CyDeploy can record and label what happens on a computer’s screen, and the machine learning comes in to learn to interpret and automatically label that activity.
To ensure there are no errors, there is still a human in the loop. A customer’s system administrator, who has the expertise to know what they are looking at or expecting, checks that the machine learning labeling is correct.
Williams-Koroma spoke ahead of the TechCrunch Disrupt 2025 conference, where CyDeploy is a Top 20 finalist in Startup Battlefield.
The advantage CyDeploy provides is that it can help write test scripts more quickly. You deploy it into your digital twin environment and have the system administrators run functional test scripts that reflect normal day-to-day uses. Then they can see from there what the issues might be.
Company customers can choose to use CyDeploy’s large language model that runs within the company’s environment to prevent proprietary data from leaving its servers. Alternatively, they can use OpenAI’s model, in which case information would go outside of the company’s systems or cloud.
Williams-Koroma said for most customers, the systems they would want to use CyDeploy on would be servers and critical machines, rather than users’ workstations. These are Tier 1 applications where changes need to be made quickly for security reasons, but not too quickly because of operational reasons.