CI/CD optimization
Designing ways to move from pipeline performance and failures into the jobs, stages, and patterns that explain where delivery time is going.
Making software delivery easier to understand and act on.
At Datadog, I work across CI/CD and software-delivery experiences: helping engineering teams move from pipeline activity and delivery signals to a clearer picture of what is slowing them down, what changed, and where to act.
Software deliveryLast 7 days
CODEPull requestReview & quality signals
→CIPipelineBuild · test · analyze
→DEPLOYProductionRelease & change signals
→LEARNDelivery healthPerformance over time
THE PROBLEM
Delivery data exists everywhere. Understanding it is the hard part.
Modern delivery workflows generate a lot of information: pipeline runs, failures, deployments, code-quality signals, pull requests, incidents, and team-level metrics. The design problem is not simply exposing more of it. It is deciding what matters in each moment and connecting those signals into a workflow developers can actually use.
My work has focused on reducing that distance between signal and action: clearer navigation, better context, stronger hierarchy, and workflows that help teams investigate without first having to understand the shape of the underlying data.
A product system, not a single screen.
Flaky test management
Helping teams identify, investigate, and manage unreliable tests so recurring failures become actionable signals instead of background noise.
DORA metrics
Turning delivery metrics into context teams can read over time, compare, and connect back to the work and changes behind them.
Software delivery
Connecting signals across the path from code to production instead of treating each part of the delivery lifecycle as an isolated tool.
AI-assisted workflows
Developing AI initiatives around log summarization and categorization, and exploring where contextual AI can reduce investigation time, surface relevant signals, and help developers move from failure to understanding and action faster.
Onboarding & settings
Making complex capabilities easier to enter and configure: clearer starting points, progressive guidance, and settings that reduce the amount a user has to learn up front.
DELIVERY HEALTHTeam performance
7d 30d 90dDeployment frequency18.4 / wk↗ steady
Lead time for changes4h 12m↘ improving
Change failure rate6.8%→ stable
Time to restore38m↘ improving
DESIGN APPROACH
Start with the question, not the dashboard.
The recurring question behind this work is: what is the user trying to understand right now? A platform can expose hundreds of dimensions, but the interface still needs to create a useful path through them.
That means designing progressive layers of information: a strong overview, a clear anomaly or signal, and then enough detail to investigate. The goal is to preserve the power of the underlying system without making complexity the user’s problem.
01
Make the next question obvious.
Every overview should create a clear path into investigation.
02
Keep context attached to the signal.
Metrics become useful when users can connect them to pipelines, changes, and teams.
03
Reveal complexity progressively.
Start readable, then let experienced users go deeper without flattening the product.
This is a portfolio-level view of the problems and design patterns I work on at Datadog. Product details and internal information are intentionally abstracted; the interface examples on this page are illustrative reconstructions rather than production screenshots.