Hi, I'm _

Some patients referred through Rula's Direct Booking Tool are Medicaid members who need to schedule with an outside partner instead of booking with Rula. This project detects those patients and routes them to clear next steps instead of dead-ending their booking.

• Added a new partner configuration field across the stack: the config parser schema, the partner service, and the partner configs themselves
• Built the external-referral instructions route, a partner registry, and a referral card component with Storybook coverage
• Added a neutral inline-notice variant to the design system and the redirect logic that sends Medicaid submissions to the partner instructions
• Worked with design to disable unsupported form options instead of hiding them, which matches existing patterns in the tool

The registry-and-redirect pattern can be reused for future external scheduling partners.

A launch site for Purpose Plate, a new social-enterprise meal company whose purchases help provide meals for people in need. The site is a funnel, not a store: every page leads visitors to the company's ordering platform, which handles menus, payments, and pickup. The site itself holds no cart and stores no customer data.

• Ran discovery before any design work began, including a project brief that flagged, early, the legal constraints on how a for-profit company can describe its give-back
• Worked alongside the client's marketing agency, turning their design prototypes into a statically generated Next.js 15 site deployed on Netlify
• Built a strict design-token layer in Tailwind v4, with tests that block utilities the design doesn't allow, and audited every color pairing for WCAG AA contrast
• Gated CI on linting, type checks, unit tests, Playwright end-to-end tests with axe accessibility checks, and Lighthouse performance budgets
• Enforced import boundaries between features so parallel AI coding agents could build separate parts of the site at the same time without colliding. That shipped 70+ merged PRs in about two months
• Wired up launch SEO (sitemap, robots, llms.txt, Open Graph cards, canonical-domain redirects) and PostHog analytics on outbound order clicks

Added a confirm-or-cancel step to Rula's appointment reminder emails and texts, so patients can tell us whether they're going 48 and 24 hours before a session. The goal was to cut no-shows among partner-referred patients. I was the sole developer and wrote the technical design myself, both firsts for me. AI tools were central to how I worked across planning, review, and implementation.

Design & planning

• Wrote the technical design document. I ran an architecture spike on where confirmation status should live and moved away from the lifecycle-messaging vendor to a webhook-based design that keeps Rula's own services as the source of truth
• Defended the choice of service in writing and answered a security review covering signed, expiring confirmation links, PHI minimization, authenticated route splitting, rate limiting, and webhook trust
• Used AI to produce a milestone-by-milestone timeline estimate, which my manager added to the initiative and used for leadership reporting
• Used AI to work through the security review's six concerns one by one. The security team accepted every response without asking for revisions

Implementation (three codebases, two of them new to me)

• Used AI pair programming to ramp up on unfamiliar codebases in hours instead of days. I merged my first patient workflow service PR within 48 hours of opening the repo, and eight PRs across two new repos within two weeks
• Patient workflow service: confirmation entity, store, service layer, and a public, rate-limited confirmation endpoint backed by DynamoDB
• Telehealth app: a feature-flagged confirmation page with appointment details and pre-appointment tasks
• Shared component library: a new wordmark component and brand color support

Launch & follow-through

• Rolled the LaunchDarkly experiment out to production, ramping from a small holdout to an even three-way split
• Resolved a production incident where the page created survey records on load instead of on click, which was skewing metrics
• Shipped the fast-follow milestones and a QA pass, then turned our retro into a cross-functional collaboration guide that the team now uses, and proposed packaging it as a Claude skill so it feeds directly into the team's AI-assisted workflows