Making machine learning readable under pressure.

IT teams running large, mixed environments were flooded with alerts. Sorting out which ones mattered was close to impossible. Behavioral Correlation used machine learning to find patterns in that noise and group related events into something people could act on.

The design challenge was making that intelligence readable. People working an outage don't have time to learn how a model works. They need to understand what it's telling them, quickly, and trust that it's real.

The model could be as complex as it needed to be. The screen couldn't be.

There was a second problem too. Six development teams were building different parts of the feature at different stages, some ahead, some behind, some changing direction. Keeping the experience consistent while all of that was moving was its own design problem.

I started with how operators already worked: the words they used, how they thought about the services they were responsible for, and what a real problem looked like to them compared with noise.

From there, the design put the model's results in those terms. Operators see their business services, how the parts connect, and plain signals like health, availability, and risk. The machine learning shows up as grouped events and anomalies tied to a service, not as model output to interpret.