AI Platform: docs assistant and engineering agents

• •Designed and built an internal RAG assistant on Postgres and pgvector that answers engineering questions from company docs through a Slack bot, a web UI, and an MCP server for coding agents. One deployment serves 1,000+ engineers and about 1,500 questions a month, with every answer scored by an LLM judge so retrieval and model changes are measured, not eyeballed.
• •Built an AI code-review agent on the company's shared model gateway that reviews pull requests across an engineering org of 1,500 and posts findings as review comments.
• •Built an AI spec-driven workflow that turns a one-line request into a plan, design doc, and task breakdown grounded in code search and internal docs; adopted by 15 teams (150 engineers).

Progressive delivery and release safety

• •Proposed and built a progressive deployment service for frontend releases: a declarative rollout.yml modeled on Argo Rollouts, a Node.js state-machine and worker engine, and automatic rollback when error rate, latency, or conversion regress.
• •Set the org standard for controlled rollouts: wrote the project plan, coordinated infra with the platform team, and moved 15 production services, one per product team, onto metric-gated releases for Credit Karma's 140M+ members.
• •Made bad releases revert themselves with a required error boundary, shared critical-error reporting, and post-publish smoke tests that gate promotion.

Frontend efficiency platform (sole lead)

• •Set the platform strategy from local development through production deployment after interviewing engineers across product teams, then mentored teams through adoption, including fast builds of embedded web apps (EWA) on physical devices and simulators.
• •Proposed and led a local development stack (one Next.js dev server, a local proxy, state presets, and MSW mocks) that lets 1,000 frontend engineers test EWA changes locally without deploying an environment, cutting time to merge from three days to one.
• •Rebuilt preview environments on production's Next.js standalone runtime, cutting preview turnaround from 30 minutes to under 5, and moved deploy configuration into a CI-validated runtime datastore so config changes ship without a redeploy. Leadership credits the platform with saving hundreds of engineer-hours a month.

Credit Karma Money

• •Led the team of 15 engineers that built Credit Karma Money, the company's first money product, on GraphQL, React, and Next.js, taking it from concept to launch in 6 months for Credit Karma's 140M+ members.

• •Built the Lyft Developer Portal, the docs and onboarding site for third-party API partners, working directly with external integration partners.
• •Developed high-traffic Lyft.com properties and led a design-system re-architecture across them.

• •Architected the frontend of a microdonation platform from the ground up, including its build, test, and deploy strategy.

• •Built React and e-commerce sites for clients including Goop, BuzzFeed, Dollar Shave Club, and LEGO.