More projects

Additional AMD work.

Projects kept off the homepage so the main page stays fast to scan, while still giving technical reviewers a deeper path.

Real-time dashboard High priority

Situation: Teams had to wait for long automated runs to finish before seeing which AI models passed, failed, or regressed.
Task: Give engineers and managers real-time visibility into model performance across daily and weekly runs.
Action: Built a Flask dashboard with live model-by-model updates, admin controls, filters, platform toggles, and automatic Jira plus email alerts for failures.
Result: Eliminated dashboard lag, improved issue visibility, and kept it responsive by sharding data and running multiple server instances.

Product engineering Public writeup

Situation: Java users needed a cleaner way to use AMD's AI acceleration work.
Task: Turn a low-level performance library into something easier for Java teams to test and adopt.
Action: Built the Java interface, examples, validation tests, user guide, and public writeup.
Result: Moved the work from internal engineering effort to a shipping, user-facing path.

Agent workflow Design contribution

Situation: Nightly and weekly performance runs could flag regressions, but developers still had to confirm whether the issue was real and where it came from.
Task: Help design a workflow that turned raw regression alerts into useful debugging context before a developer opened the ticket.
Action: Contributed to the project design by mapping the triage loop: rerun tests, check reproducibility, analyze likely causes, and post findings back to Jira.
Result: Helped shape a concrete agent workflow that could give developers a clearer starting point for confirmed performance regressions.