Staff Software Engineer
& Team Lead

The problem: Data workers across the enterprise were abandoning search queries and asking colleagues on Slack instead. Queries took 3-5 seconds. Precision was poor. Product identified it as a top pain point.

What was built: Worked with UX designers in Figma and backend engineers to define clean API contracts and module boundaries. Designed a component-based frontend architecture with a thin TypeScript client talking to the Search Service REST API. The core decision: make the entire faceted search UI config-driven. Each facet type — dropdown, checkbox, date range, autocomplete — is defined in JSON configuration. Adding a new filter means adding a config entry, not writing new code. That cost two extra sprints upfront but paid for itself within the first quarter as 10+ new filter types were added through config alone.

The result: Queries dropped from 3-5 seconds to 200-400ms. Set up CI/CD for 2 daily releases and maintained 99.9% uptime processing 1M+ daily requests. Two of the three engineers mentored were promoted within a year.

Part of Intuit's Data Mesh initiative, featured at AWS re:Invent and in Intuit's engineering blog.

The problem: When users opened large data lineage graphs (2,500+ nodes), the app froze for up to 3 minutes. Memory spiked to 700MB. People were force-quitting browsers. 500+ users were affected daily, losing roughly 2 hours of collective productivity.

The wrong assumption: Initially assumed the backend API was the bottleneck — the data volumes felt huge. Spent two days profiling the wrong layer and coordinating with the backend team. The backend was fine.

The real fix: Chrome DevTools Network tab revealed the real issues: 3 API calls when only 2 were needed (a leftover from an earlier implementation), all 2,500+ rows rendered in the DOM without virtualization, and state duplicated across components. Each one was bad; together they were catastrophic on large graphs. Eliminated the redundant call, parallelized the remaining two, migrated to React Base Table with virtual scrolling, introduced a shared Redux store, added async data processing, and optimized component lifecycles. Wrote 24 regression tests and stress-tested with the largest production graphs.

The lesson: Profile the full request chain before assuming anything. Let the data tell you where the problem is. This became the team's documented best practice for performance debugging.

The problem: Frontend technical debt was invisible across the org. No consistent way to measure RUM adoption, code coverage, test infrastructure, or documentation across 13 plugins and multiple teams. Teams couldn't advocate for remediation time because they couldn't quantify the problem. Several frontend leads pushed back hard — they saw this as extra work that would just expose gaps.

What was built: A phased scorecard framework. Phase 1 established baselines without requiring immediate fixes — no team refused, because nothing was being asked of them yet except measurement. Phase 2 introduced prioritized targets. Built quick prototypes to make conversations concrete. Framed it as a tool that helps teams advocate for dedicated tech debt time on their roadmaps, not a stick to beat them with. That reframing was the turning point.

The result: 80% coverage across all frontend plugins. $10K+ in OpEx savings identified from fewer production incidents. Teams that were initially resistant adopted the framework, and the director asked for the cross-org excellence work to continue into the next quarter.

The problem: A 10-person frontend team under pressure to ship more per quarter, but velocity had plateaued. More sprints and more hours weren't sustainable. Nobody had demonstrated concrete AI gains yet.

What was done: Proved it with real output first — integrated Cursor AI into daily workflow for scaffolding, test generation, refactoring, and PR descriptions. Tracked results: consistently hitting 2-3x coding speed on UI tasks. Scaffolding a React component with tests and Storybook went from ~45 minutes to ~15-20 minutes. Hosted a brown bag using real backlog tasks, not toy examples, with 90% attendance. Presented at an engineering-wide forum for 50+ attendees. Enabled a PR review agent across repositories that caught missing null checks, inconsistent error handling, and unused imports — so human reviewers could focus on design and logic.

The result: Team-wide adoption across the broader 50-person org. PRs passing review at the same quality bar, just faster. Manager noted the team was "concretely accelerating the way we deliver through AI tooling."