The fastest way to know if something works is to

At Lirio, the content team produces health intervention content for multiple channels and languages. Each with its own constraints and rendering rules.

The team works on tight deadlines, authoring content and scoring it against behavioral science principles before it's put into production so it has the best chance of success.

The content team had to use past experience, approximated mockups, and assumptions while they were writing content. This was risky and time-consuming, and lead to truncated text, broken rendering, blown SMS segment limits, and costly downstream revisions.

New requirements kept raising the stakes: Japanese-language push notifications, voice messages, RCS complexity, the WhatsApp launch, and MyChart integration.

I built a reusable set of React UI components (documented, accessible, and consistent via Storybook) so previews looked like the real thing.

Building on a Figma mock-up system I'd created years earlier, I developed an interactive message preview tool for the content team covering every channel: email, SMS (with segment-aware optimization), RCS simulations, pixel-perfect WhatsApp and MyChart renderings (inbox and detail views), and iOS/Android push notifications optimized against platform truncation limits.

I added a voice preview feature using Microsoft text-to-speech, and integrated an LLM via Microsoft Foundry so the team could author base content, tailor it per channel, and generate language variations with a single click.

I shipped the push notification preview first to prove value early, then added an export function for sharing client-ready mockups. I deliberately prioritized getting a functional tool into the team's hands over visual polish.

• Caught truncation and rendering issues before launch, cutting content errors and rework.
• Reduced send costs through SMS segment optimization and one-click multi-language generation.
• Brought previews into the design system as a reusable, documented React component library.
• Shipped a multi-channel preview tool spanning email, SMS, RCS, WhatsApp, MyChart, push, and voice.
• Streamlined client sharing through a consistent mockup export function.

I wanted to see if I could build a real product for myself during a week off work, using Claude Code as my primary collaborator.

The real question wasn't whether AI could generate code. It was whether I, as a designer, could use AI-assisted tools to ship an actual product end to end, making the continuous stream of small decisions that separate a one-shot prompt from something real.

I also wanted to pressure-test where AI genuinely helps in the design and build loop versus where human judgment still has to do the heavy lifting.

I planned the app and let AI handle most of the typing: owning the requirements, iterating until it felt right, and deciding what shipped or went in the bin. I wired together ~10 services (Next.js, Supabase, Resend, Claude Haiku, Vercel cron, Mapbox, Open-Meteo, and a few content APIs) across 441+ commits and 188+ merged PRs over ~70 days.

I designed the Figma file roughly 60% by hand and 40% with AI (Claude Code plus the Figma MCP server), using AI to brainstorm landing-page layouts and explore color palettes for unified-yet-distinct icons, while keeping components, the final palette, onboarding, and shipped design under my own hand.

I engineered a Southern voice through a system prompt built from description and examples, and learned in production to limit rewrites to the preheader, weather, and horoscope after rewriting the joke and riddle ruined them. I built a contrast-aware procedural pipeline for quote backgrounds (mood classification via Claude Haiku, luminosity-sampled contrast strips, auto text-color selection, Satori/Sharp rendering to a 600×315 PNG), and explored a generative-image version.

Most importantly, I built a per-user news pipeline in real code, then deleted it, because the voice and curation were what made Rooster distinctive and commodity news wasn't.

• Proved a designer can own the full loop: plan, iterate, and decide what ships, with AI as a design material rather than a replacement for judgment.
• Designed, built, and shipped a real, working product solo; the content pipeline came together in a single day, with real emails landing the next morning.
• Sustained delivery over ~70 days: 441+ commits, 188+ merged PRs, ~10 services in production.
• Showed the value of deleting a feature: AI made building cheap enough to afford being wrong and changing course.
• Turned the work into a talk on how AI collapses the wall between designing and building.

Lirio used behavioral science content to help patients overcome barriers to scheduling critical health appointments (mammograms, colonoscopies, vaccinations).

The content powering these interventions was managed in massive spreadsheets (sometimes 50+ columns wide) duplicated and modified for every client's brand, compliance, and legal requirements. This same content fed the AI system that selected and composed personalized messages.

The spreadsheet process took two to three months to get a client's content approved, and most contracts prevented billing until messages were actively sending, so the process was directly blocking revenue.

Structural errors in the spreadsheets degraded AI message quality, inconsistent tags made personalization unreliable, client feedback had no tracking mechanism, and messaging errors still reached production.

As principal and sole product designer, I led the design of Lirio's first user-facing product: a custom CMS built for behavioral science and optimized for AI-driven content selection.

Rather than digitizing the existing spreadsheet, I mapped the full content lifecycle by interviewing behavioral design, AI engineering, platform engineering, and client onboarding teams. I owned user research, information architecture and content modeling, workflow and UI design, high-fidelity prototyping, and engineering implementation guidance.

I designed a structured content model aligned with AI interventions, built-in review and approval workflows with audit trails, live email and SMS previews, role-based collaboration, activation-pipeline integration, and reusable components that seeded a scalable design system. Discovery to production MVP took four months (March to July 2020).

• Reduced client approval time by 75%.
• Reduced content creation time by more than 50%.
• Unlocked earlier revenue by accelerating message activation.
• Eliminated messaging and composition errors that previously reached production.
• Strengthened AI content-selection reliability by cleaning up source structure.
• Created a scalable foundation for future intervention tooling and the design system.

Watershed launched a new category of learning analytics platform built on xAPI, an emerging data standard that allowed far richer, more flexible tracking of corporate training than the incumbent SCORM.

The market was still constrained by SCORM and couldn't yet imagine what modern learning analytics could do. When I joined, the product was barely more than a login screen and a raw activity log of xAPI statements.

Two challenges at once: design a fully functional analytics platform from near-scratch, and create a compelling story that would help buyers understand why xAPI mattered.

There were no modern tools for exploring learning data, and the category itself had to be evangelized before it could be sold.

I served as both Principal Product Designer and Product Owner. I led user research and requirements definition, designed and launched 15 customizable report types, developed a data visualization strategy tailored to L&D, built frontend components and dashboard configuration tools, and coordinated with 8 engineers from concept to production.

I designed the system end to end: report architecture, dashboard structure, sharing and configuration flows, data ingestion and governance tooling, and admin tools. I created domain-specific visualizations (Activity, Outcomes, Leaderboard, Matching reports) and built market-facing demos and concepts that sales used to win clients before the full feature set existed.

I also shaped the product narratives used directly in conversations with Google, AT&T, and Visa.

• Contributed to a successful exit: Watershed was acquired by Learning Technologies Group (LTG).
• Helped close pilot deals with Google, AT&T, and Visa.
• Played a key role in defining and evangelizing the xAPI standard.
• Designed and launched a fully customizable learning analytics platform.
• Many tools I designed replaced traditional BI dashboards in L&D organizations.
• Contributed to industry-wide adoption of learning analytics platforms.

Emma, an email marketing company, wanted to extend its platform into mobile. As mobile lead, I worked with the product leadership team to create an iPad app (Guestbook) that businesses could use to sign people up for their email lists, running on an iPad at a counter or carried around at events.

The catch: Emma had no available iOS developers or designers, and core internal API functionality didn't yet exist to support the app.

The product had to be built essentially from zero, with no in-house mobile talent, no backend capabilities, and no validated understanding of how businesses would actually use an in-person signup tool.

Everything from team assembly to API specification to budget had to be stood up alongside the design work.

I owned product management, user research, information architecture, interaction design, prototyping, and QA.

I interviewed local businesses to understand real-world use cases, then sketched the needed capabilities and built a native coded prototype on an actual iPad, because I wanted to validate the design on real hardware, not in a mockup. I ran informal usability sessions by carrying the iPad around the office, which drove several design refinements.

I recruited and selected a contract iOS developer and mobile designer, using the prototype to align them and inform their quotes. I coordinated with Emma's API team to specify and develop the required backend, the finance team to set up the App Store account, secure budget, and pay vendors, and the marketing team to produce an intro video and App Store copy. I QA'd the app across current and previous-generation iPads before launch.

• Launched Emma's first iPad app (Guestbook) end to end, from research through App Store release.
• Stood up mobile capability for a company with no existing iOS team, assembling vendors, budget, and backend support.
• Saw strong positive feedback at launch, with Emma customers adopting it immediately to grow their email lists.
• Validated the design through hands-on prototyping and usability testing on real hardware.

This was my first major project for Emma, an email marketing company. As Product Owner and Lead Designer, I was tasked with building the email market's first interactive drag-and-drop email editor, a category-defining product.

I led a team of five designers and set requirements and priorities for both internal and external development teams.

The existing email editing experience had clear limitations, and customers wanted capabilities they couldn't yet access.

Creating an interactive drag-and-drop editor meant designing new interaction patterns from scratch, while coordinating multiple development teams and a separate workflow for Emma's design team to upload and manage customer templates.

I owned product management, user research, information architecture, interaction design, and QA, and served as product owner across cross-functional stakeholders.

I started with research, forming a core team of internal stakeholders representing different Emma customer types, and learning both what customers loved about the current editor and what they wished they could do. I led several design sprints with Emma's other UX designers to generate interaction concepts, then created mockups of the core workflow the team built from.

After gathering estimates, I created the product roadmap, defined feature sets, and planned development sprints. I also designed the separate template upload-and-management workflow and handed those mockups to a different internal dev team to estimate and build.

Throughout, I shared progress in weekly staff meetings using short videos to show rather than tell, which drew strong, enthusiastic responses.

• Led design and product ownership for the email market's first interactive drag-and-drop email editor.
• Directed a team of five designers and coordinated multiple internal and external development teams.
• Delivered a roadmap, defined feature sets, and sprint plans that moved the editor from concept to build.
• Grounded the product in research with stakeholders representing Emma's full customer range.
• Built organizational momentum through progress videos that generated visible enthusiasm across the company.